from __future__ import annotations
import torch
class GraphModule(torch.nn.Module):
    def forward(self, s59: "Sym(s59)", L_inputs_embeds_: "bf16[s59, 5120]", s72: "Sym(s59)", L_input_ids_: "i32[s59]", L_positions_: "i64[s59]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_: "bf16[1280]", L_self_modules_audio_language_adapter_modules_w_in_parameters_weight_: "bf16[3072, 5120]", L_self_modules_audio_language_adapter_modules_w_out_parameters_weight_: "bf16[3072, 3072]", L_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_: "bf16[131072, 3072]", L_self_modules_time_embedding_buffers_inv_freq_: "f32[1536]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", L_self_modules_language_model_modules_model_modules_norm_parameters_weight_: "bf16[3072]"):
        l_inputs_embeds_ = L_inputs_embeds_
        l_input_ids_ = L_input_ids_
        l_positions_ = L_positions_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_ = L_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_
        l_self_modules_audio_language_adapter_modules_w_in_parameters_weight_ = L_self_modules_audio_language_adapter_modules_w_in_parameters_weight_
        l_self_modules_audio_language_adapter_modules_w_out_parameters_weight_ = L_self_modules_audio_language_adapter_modules_w_out_parameters_weight_
        l_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_ = L_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_
        l_self_modules_time_embedding_buffers_inv_freq_ = L_self_modules_time_embedding_buffers_inv_freq_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_ = L_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_
        l_self_modules_language_model_modules_model_modules_norm_parameters_weight_ = L_self_modules_language_model_modules_model_modules_norm_parameters_weight_
        
        # No stacktrace found for following nodes
        submod_0 = self.submod_0(s59, l_inputs_embeds_, l_positions_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  l_inputs_embeds_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem = submod_0[0]
        getitem_1 = submod_0[1]
        getitem_2 = submod_0[2]
        getitem_3 = submod_0[3]
        getitem_4 = submod_0[4]
        getitem_5 = submod_0[5]
        getitem_6 = submod_0[6];  submod_0 = None
        submod_1 = self.submod_1(getitem, s59, getitem_1, getitem_2, getitem_3);  getitem = getitem_1 = getitem_2 = submod_1 = None
        submod_2 = self.submod_2(getitem_3, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_, getitem_4, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_4 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_7 = submod_2[0]
        getitem_8 = submod_2[1]
        getitem_9 = submod_2[2]
        getitem_10 = submod_2[3]
        getitem_11 = submod_2[4];  submod_2 = None
        submod_3 = self.submod_3(getitem_7, s59, getitem_8, getitem_9, getitem_10);  getitem_7 = getitem_8 = getitem_9 = submod_3 = None
        submod_4 = self.submod_4(getitem_10, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_, getitem_11, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_10 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_11 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_12 = submod_4[0]
        getitem_13 = submod_4[1]
        getitem_14 = submod_4[2]
        getitem_15 = submod_4[3]
        getitem_16 = submod_4[4];  submod_4 = None
        submod_5 = self.submod_5(getitem_12, s59, getitem_13, getitem_14, getitem_15);  getitem_12 = getitem_13 = getitem_14 = submod_5 = None
        submod_6 = self.submod_6(getitem_15, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_, getitem_16, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_15 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_16 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_17 = submod_6[0]
        getitem_18 = submod_6[1]
        getitem_19 = submod_6[2]
        getitem_20 = submod_6[3]
        getitem_21 = submod_6[4];  submod_6 = None
        submod_7 = self.submod_7(getitem_17, s59, getitem_18, getitem_19, getitem_20);  getitem_17 = getitem_18 = getitem_19 = submod_7 = None
        submod_8 = self.submod_8(getitem_20, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_, getitem_21, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_20 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_21 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_22 = submod_8[0]
        getitem_23 = submod_8[1]
        getitem_24 = submod_8[2]
        getitem_25 = submod_8[3]
        getitem_26 = submod_8[4];  submod_8 = None
        submod_9 = self.submod_9(getitem_22, s59, getitem_23, getitem_24, getitem_25);  getitem_22 = getitem_23 = getitem_24 = submod_9 = None
        submod_10 = self.submod_10(getitem_25, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_, getitem_26, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_25 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_26 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_27 = submod_10[0]
        getitem_28 = submod_10[1]
        getitem_29 = submod_10[2]
        getitem_30 = submod_10[3]
        getitem_31 = submod_10[4];  submod_10 = None
        submod_11 = self.submod_11(getitem_27, s59, getitem_28, getitem_29, getitem_30);  getitem_27 = getitem_28 = getitem_29 = submod_11 = None
        submod_12 = self.submod_12(getitem_30, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_, getitem_31, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_30 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_31 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_32 = submod_12[0]
        getitem_33 = submod_12[1]
        getitem_34 = submod_12[2]
        getitem_35 = submod_12[3]
        getitem_36 = submod_12[4];  submod_12 = None
        submod_13 = self.submod_13(getitem_32, s59, getitem_33, getitem_34, getitem_35);  getitem_32 = getitem_33 = getitem_34 = submod_13 = None
        submod_14 = self.submod_14(getitem_35, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_, getitem_36, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_35 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_36 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_37 = submod_14[0]
        getitem_38 = submod_14[1]
        getitem_39 = submod_14[2]
        getitem_40 = submod_14[3]
        getitem_41 = submod_14[4];  submod_14 = None
        submod_15 = self.submod_15(getitem_37, s59, getitem_38, getitem_39, getitem_40);  getitem_37 = getitem_38 = getitem_39 = submod_15 = None
        submod_16 = self.submod_16(getitem_40, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_, getitem_41, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_40 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_41 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_42 = submod_16[0]
        getitem_43 = submod_16[1]
        getitem_44 = submod_16[2]
        getitem_45 = submod_16[3]
        getitem_46 = submod_16[4];  submod_16 = None
        submod_17 = self.submod_17(getitem_42, s59, getitem_43, getitem_44, getitem_45);  getitem_42 = getitem_43 = getitem_44 = submod_17 = None
        submod_18 = self.submod_18(getitem_45, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_, getitem_46, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_45 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_46 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_47 = submod_18[0]
        getitem_48 = submod_18[1]
        getitem_49 = submod_18[2]
        getitem_50 = submod_18[3]
        getitem_51 = submod_18[4];  submod_18 = None
        submod_19 = self.submod_19(getitem_47, s59, getitem_48, getitem_49, getitem_50);  getitem_47 = getitem_48 = getitem_49 = submod_19 = None
        submod_20 = self.submod_20(getitem_50, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_, getitem_51, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_50 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_51 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_52 = submod_20[0]
        getitem_53 = submod_20[1]
        getitem_54 = submod_20[2]
        getitem_55 = submod_20[3]
        getitem_56 = submod_20[4];  submod_20 = None
        submod_21 = self.submod_21(getitem_52, s59, getitem_53, getitem_54, getitem_55);  getitem_52 = getitem_53 = getitem_54 = submod_21 = None
        submod_22 = self.submod_22(getitem_55, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_, getitem_56, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_55 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_56 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_57 = submod_22[0]
        getitem_58 = submod_22[1]
        getitem_59 = submod_22[2]
        getitem_60 = submod_22[3]
        getitem_61 = submod_22[4];  submod_22 = None
        submod_23 = self.submod_23(getitem_57, s59, getitem_58, getitem_59, getitem_60);  getitem_57 = getitem_58 = getitem_59 = submod_23 = None
        submod_24 = self.submod_24(getitem_60, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_, getitem_61, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_60 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_61 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_62 = submod_24[0]
        getitem_63 = submod_24[1]
        getitem_64 = submod_24[2]
        getitem_65 = submod_24[3]
        getitem_66 = submod_24[4];  submod_24 = None
        submod_25 = self.submod_25(getitem_62, s59, getitem_63, getitem_64, getitem_65);  getitem_62 = getitem_63 = getitem_64 = submod_25 = None
        submod_26 = self.submod_26(getitem_65, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_, getitem_66, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_65 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_66 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_67 = submod_26[0]
        getitem_68 = submod_26[1]
        getitem_69 = submod_26[2]
        getitem_70 = submod_26[3]
        getitem_71 = submod_26[4];  submod_26 = None
        submod_27 = self.submod_27(getitem_67, s59, getitem_68, getitem_69, getitem_70);  getitem_67 = getitem_68 = getitem_69 = submod_27 = None
        submod_28 = self.submod_28(getitem_70, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_, getitem_71, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_70 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_71 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_72 = submod_28[0]
        getitem_73 = submod_28[1]
        getitem_74 = submod_28[2]
        getitem_75 = submod_28[3]
        getitem_76 = submod_28[4];  submod_28 = None
        submod_29 = self.submod_29(getitem_72, s59, getitem_73, getitem_74, getitem_75);  getitem_72 = getitem_73 = getitem_74 = submod_29 = None
        submod_30 = self.submod_30(getitem_75, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_, getitem_76, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_75 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_76 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_77 = submod_30[0]
        getitem_78 = submod_30[1]
        getitem_79 = submod_30[2]
        getitem_80 = submod_30[3]
        getitem_81 = submod_30[4];  submod_30 = None
        submod_31 = self.submod_31(getitem_77, s59, getitem_78, getitem_79, getitem_80);  getitem_77 = getitem_78 = getitem_79 = submod_31 = None
        submod_32 = self.submod_32(getitem_80, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_, getitem_81, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_80 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_81 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_82 = submod_32[0]
        getitem_83 = submod_32[1]
        getitem_84 = submod_32[2]
        getitem_85 = submod_32[3]
        getitem_86 = submod_32[4];  submod_32 = None
        submod_33 = self.submod_33(getitem_82, s59, getitem_83, getitem_84, getitem_85);  getitem_82 = getitem_83 = getitem_84 = submod_33 = None
        submod_34 = self.submod_34(getitem_85, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_, getitem_86, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_85 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_86 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_87 = submod_34[0]
        getitem_88 = submod_34[1]
        getitem_89 = submod_34[2]
        getitem_90 = submod_34[3]
        getitem_91 = submod_34[4];  submod_34 = None
        submod_35 = self.submod_35(getitem_87, s59, getitem_88, getitem_89, getitem_90);  getitem_87 = getitem_88 = getitem_89 = submod_35 = None
        submod_36 = self.submod_36(getitem_90, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_, getitem_91, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_90 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_91 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_92 = submod_36[0]
        getitem_93 = submod_36[1]
        getitem_94 = submod_36[2]
        getitem_95 = submod_36[3]
        getitem_96 = submod_36[4];  submod_36 = None
        submod_37 = self.submod_37(getitem_92, s59, getitem_93, getitem_94, getitem_95);  getitem_92 = getitem_93 = getitem_94 = submod_37 = None
        submod_38 = self.submod_38(getitem_95, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_, getitem_96, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_95 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_96 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_97 = submod_38[0]
        getitem_98 = submod_38[1]
        getitem_99 = submod_38[2]
        getitem_100 = submod_38[3]
        getitem_101 = submod_38[4];  submod_38 = None
        submod_39 = self.submod_39(getitem_97, s59, getitem_98, getitem_99, getitem_100);  getitem_97 = getitem_98 = getitem_99 = submod_39 = None
        submod_40 = self.submod_40(getitem_100, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_, getitem_101, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_100 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_101 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_102 = submod_40[0]
        getitem_103 = submod_40[1]
        getitem_104 = submod_40[2]
        getitem_105 = submod_40[3]
        getitem_106 = submod_40[4];  submod_40 = None
        submod_41 = self.submod_41(getitem_102, s59, getitem_103, getitem_104, getitem_105);  getitem_102 = getitem_103 = getitem_104 = submod_41 = None
        submod_42 = self.submod_42(getitem_105, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_, getitem_106, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_105 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_106 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_107 = submod_42[0]
        getitem_108 = submod_42[1]
        getitem_109 = submod_42[2]
        getitem_110 = submod_42[3]
        getitem_111 = submod_42[4];  submod_42 = None
        submod_43 = self.submod_43(getitem_107, s59, getitem_108, getitem_109, getitem_110);  getitem_107 = getitem_108 = getitem_109 = submod_43 = None
        submod_44 = self.submod_44(getitem_110, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_, getitem_111, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_110 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_111 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_112 = submod_44[0]
        getitem_113 = submod_44[1]
        getitem_114 = submod_44[2]
        getitem_115 = submod_44[3]
        getitem_116 = submod_44[4];  submod_44 = None
        submod_45 = self.submod_45(getitem_112, s59, getitem_113, getitem_114, getitem_115);  getitem_112 = getitem_113 = getitem_114 = submod_45 = None
        submod_46 = self.submod_46(getitem_115, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_, getitem_116, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_115 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_116 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_117 = submod_46[0]
        getitem_118 = submod_46[1]
        getitem_119 = submod_46[2]
        getitem_120 = submod_46[3]
        getitem_121 = submod_46[4];  submod_46 = None
        submod_47 = self.submod_47(getitem_117, s59, getitem_118, getitem_119, getitem_120);  getitem_117 = getitem_118 = getitem_119 = submod_47 = None
        submod_48 = self.submod_48(getitem_120, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_, getitem_121, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_120 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_121 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_122 = submod_48[0]
        getitem_123 = submod_48[1]
        getitem_124 = submod_48[2]
        getitem_125 = submod_48[3]
        getitem_126 = submod_48[4];  submod_48 = None
        submod_49 = self.submod_49(getitem_122, s59, getitem_123, getitem_124, getitem_125);  getitem_122 = getitem_123 = getitem_124 = submod_49 = None
        submod_50 = self.submod_50(getitem_125, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_, getitem_126, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_125 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_126 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_127 = submod_50[0]
        getitem_128 = submod_50[1]
        getitem_129 = submod_50[2]
        getitem_130 = submod_50[3]
        getitem_131 = submod_50[4];  submod_50 = None
        submod_51 = self.submod_51(getitem_127, s59, getitem_128, getitem_129, getitem_130);  getitem_127 = getitem_128 = getitem_129 = submod_51 = None
        submod_52 = self.submod_52(getitem_130, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_, getitem_131, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_130 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_131 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_132 = submod_52[0]
        getitem_133 = submod_52[1]
        getitem_134 = submod_52[2]
        getitem_135 = submod_52[3]
        getitem_136 = submod_52[4];  submod_52 = None
        submod_53 = self.submod_53(getitem_132, s59, getitem_133, getitem_134, getitem_135);  getitem_132 = getitem_133 = getitem_134 = submod_53 = None
        submod_54 = self.submod_54(getitem_135, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_, getitem_136, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_135 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_136 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_137 = submod_54[0]
        getitem_138 = submod_54[1]
        getitem_139 = submod_54[2]
        getitem_140 = submod_54[3]
        getitem_141 = submod_54[4];  submod_54 = None
        submod_55 = self.submod_55(getitem_137, s59, getitem_138, getitem_139, getitem_140);  getitem_137 = getitem_138 = getitem_139 = submod_55 = None
        submod_56 = self.submod_56(getitem_140, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_, getitem_141, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_140 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_141 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_142 = submod_56[0]
        getitem_143 = submod_56[1]
        getitem_144 = submod_56[2]
        getitem_145 = submod_56[3]
        getitem_146 = submod_56[4];  submod_56 = None
        submod_57 = self.submod_57(getitem_142, s59, getitem_143, getitem_144, getitem_145);  getitem_142 = getitem_143 = getitem_144 = submod_57 = None
        submod_58 = self.submod_58(getitem_145, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_, getitem_146, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_145 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_146 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_147 = submod_58[0]
        getitem_148 = submod_58[1]
        getitem_149 = submod_58[2]
        getitem_150 = submod_58[3]
        getitem_151 = submod_58[4];  submod_58 = None
        submod_59 = self.submod_59(getitem_147, s59, getitem_148, getitem_149, getitem_150);  getitem_147 = getitem_148 = getitem_149 = submod_59 = None
        submod_60 = self.submod_60(getitem_150, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_, getitem_151, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_150 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_151 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
        getitem_152 = submod_60[0]
        getitem_153 = submod_60[1]
        getitem_154 = submod_60[2]
        getitem_155 = submod_60[3]
        getitem_156 = submod_60[4];  submod_60 = None
        submod_61 = self.submod_61(getitem_152, s59, getitem_153, getitem_154, getitem_155);  getitem_152 = getitem_153 = getitem_154 = submod_61 = None
        submod_62 = self.submod_62(getitem_155, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_, getitem_156, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_, getitem_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_, getitem_6);  getitem_155 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_156 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_ = getitem_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = getitem_6 = None
        getitem_157 = submod_62[0]
        getitem_158 = submod_62[1]
        getitem_159 = submod_62[2]
        getitem_160 = submod_62[3]
        getitem_161 = submod_62[4];  submod_62 = None
        submod_63 = self.submod_63(getitem_157, s59, getitem_158, getitem_159, getitem_160);  getitem_157 = getitem_158 = getitem_159 = submod_63 = None
        submod_64 = self.submod_64(getitem_160, s59, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_, getitem_161, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_, l_self_modules_audio_language_adapter_modules_w_in_parameters_weight_, l_self_modules_audio_language_adapter_modules_w_out_parameters_weight_, l_input_ids_, l_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_, l_self_modules_time_embedding_buffers_inv_freq_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_160 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_ = getitem_161 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_ = l_self_modules_audio_language_adapter_modules_w_in_parameters_weight_ = l_self_modules_audio_language_adapter_modules_w_out_parameters_weight_ = l_input_ids_ = l_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_ = l_self_modules_time_embedding_buffers_inv_freq_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_162 = submod_64[0]
        getitem_163 = submod_64[1]
        getitem_164 = submod_64[2]
        getitem_165 = submod_64[3]
        getitem_166 = submod_64[4]
        getitem_167 = submod_64[5];  submod_64 = None
        submod_65 = self.submod_65(getitem_162, s59, getitem_163, getitem_164, getitem_165);  getitem_162 = getitem_163 = getitem_164 = submod_65 = None
        submod_66 = self.submod_66(getitem_165, s59, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_, getitem_166, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_165 = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_ = getitem_166 = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_168 = submod_66[0]
        getitem_169 = submod_66[1]
        getitem_170 = submod_66[2]
        getitem_171 = submod_66[3]
        getitem_172 = submod_66[4];  submod_66 = None
        submod_67 = self.submod_67(getitem_168, s59, getitem_169, getitem_170, getitem_171);  getitem_168 = getitem_169 = getitem_170 = submod_67 = None
        submod_68 = self.submod_68(getitem_171, s59, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_, getitem_172, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_171 = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_ = getitem_172 = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_173 = submod_68[0]
        getitem_174 = submod_68[1]
        getitem_175 = submod_68[2]
        getitem_176 = submod_68[3]
        getitem_177 = submod_68[4];  submod_68 = None
        submod_69 = self.submod_69(getitem_173, s59, getitem_174, getitem_175, getitem_176);  getitem_173 = getitem_174 = getitem_175 = submod_69 = None
        submod_70 = self.submod_70(getitem_176, s59, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_, getitem_177, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_176 = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_ = getitem_177 = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_178 = submod_70[0]
        getitem_179 = submod_70[1]
        getitem_180 = submod_70[2]
        getitem_181 = submod_70[3]
        getitem_182 = submod_70[4];  submod_70 = None
        submod_71 = self.submod_71(getitem_178, s59, getitem_179, getitem_180, getitem_181);  getitem_178 = getitem_179 = getitem_180 = submod_71 = None
        submod_72 = self.submod_72(getitem_181, s59, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_, getitem_182, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_181 = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_ = getitem_182 = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_183 = submod_72[0]
        getitem_184 = submod_72[1]
        getitem_185 = submod_72[2]
        getitem_186 = submod_72[3]
        getitem_187 = submod_72[4];  submod_72 = None
        submod_73 = self.submod_73(getitem_183, s59, getitem_184, getitem_185, getitem_186);  getitem_183 = getitem_184 = getitem_185 = submod_73 = None
        submod_74 = self.submod_74(getitem_186, s59, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_, getitem_187, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_186 = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_ = getitem_187 = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_188 = submod_74[0]
        getitem_189 = submod_74[1]
        getitem_190 = submod_74[2]
        getitem_191 = submod_74[3]
        getitem_192 = submod_74[4];  submod_74 = None
        submod_75 = self.submod_75(getitem_188, s59, getitem_189, getitem_190, getitem_191);  getitem_188 = getitem_189 = getitem_190 = submod_75 = None
        submod_76 = self.submod_76(getitem_191, s59, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_, getitem_192, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_191 = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_ = getitem_192 = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_193 = submod_76[0]
        getitem_194 = submod_76[1]
        getitem_195 = submod_76[2]
        getitem_196 = submod_76[3]
        getitem_197 = submod_76[4];  submod_76 = None
        submod_77 = self.submod_77(getitem_193, s59, getitem_194, getitem_195, getitem_196);  getitem_193 = getitem_194 = getitem_195 = submod_77 = None
        submod_78 = self.submod_78(getitem_196, s59, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_, getitem_197, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_196 = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_ = getitem_197 = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_198 = submod_78[0]
        getitem_199 = submod_78[1]
        getitem_200 = submod_78[2]
        getitem_201 = submod_78[3]
        getitem_202 = submod_78[4];  submod_78 = None
        submod_79 = self.submod_79(getitem_198, s59, getitem_199, getitem_200, getitem_201);  getitem_198 = getitem_199 = getitem_200 = submod_79 = None
        submod_80 = self.submod_80(getitem_201, s59, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_, getitem_202, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_201 = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_ = getitem_202 = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_203 = submod_80[0]
        getitem_204 = submod_80[1]
        getitem_205 = submod_80[2]
        getitem_206 = submod_80[3]
        getitem_207 = submod_80[4];  submod_80 = None
        submod_81 = self.submod_81(getitem_203, s59, getitem_204, getitem_205, getitem_206);  getitem_203 = getitem_204 = getitem_205 = submod_81 = None
        submod_82 = self.submod_82(getitem_206, s59, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_, getitem_207, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_206 = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_ = getitem_207 = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_208 = submod_82[0]
        getitem_209 = submod_82[1]
        getitem_210 = submod_82[2]
        getitem_211 = submod_82[3]
        getitem_212 = submod_82[4];  submod_82 = None
        submod_83 = self.submod_83(getitem_208, s59, getitem_209, getitem_210, getitem_211);  getitem_208 = getitem_209 = getitem_210 = submod_83 = None
        submod_84 = self.submod_84(getitem_211, s59, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_, getitem_212, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_211 = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_ = getitem_212 = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_213 = submod_84[0]
        getitem_214 = submod_84[1]
        getitem_215 = submod_84[2]
        getitem_216 = submod_84[3]
        getitem_217 = submod_84[4];  submod_84 = None
        submod_85 = self.submod_85(getitem_213, s59, getitem_214, getitem_215, getitem_216);  getitem_213 = getitem_214 = getitem_215 = submod_85 = None
        submod_86 = self.submod_86(getitem_216, s59, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_, getitem_217, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_216 = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_ = getitem_217 = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_218 = submod_86[0]
        getitem_219 = submod_86[1]
        getitem_220 = submod_86[2]
        getitem_221 = submod_86[3]
        getitem_222 = submod_86[4];  submod_86 = None
        submod_87 = self.submod_87(getitem_218, s59, getitem_219, getitem_220, getitem_221);  getitem_218 = getitem_219 = getitem_220 = submod_87 = None
        submod_88 = self.submod_88(getitem_221, s59, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_, getitem_222, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_221 = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_ = getitem_222 = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_223 = submod_88[0]
        getitem_224 = submod_88[1]
        getitem_225 = submod_88[2]
        getitem_226 = submod_88[3]
        getitem_227 = submod_88[4];  submod_88 = None
        submod_89 = self.submod_89(getitem_223, s59, getitem_224, getitem_225, getitem_226);  getitem_223 = getitem_224 = getitem_225 = submod_89 = None
        submod_90 = self.submod_90(getitem_226, s59, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_, getitem_227, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_226 = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_ = getitem_227 = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_228 = submod_90[0]
        getitem_229 = submod_90[1]
        getitem_230 = submod_90[2]
        getitem_231 = submod_90[3]
        getitem_232 = submod_90[4];  submod_90 = None
        submod_91 = self.submod_91(getitem_228, s59, getitem_229, getitem_230, getitem_231);  getitem_228 = getitem_229 = getitem_230 = submod_91 = None
        submod_92 = self.submod_92(getitem_231, s59, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_, getitem_232, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_231 = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_ = getitem_232 = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_233 = submod_92[0]
        getitem_234 = submod_92[1]
        getitem_235 = submod_92[2]
        getitem_236 = submod_92[3]
        getitem_237 = submod_92[4];  submod_92 = None
        submod_93 = self.submod_93(getitem_233, s59, getitem_234, getitem_235, getitem_236);  getitem_233 = getitem_234 = getitem_235 = submod_93 = None
        submod_94 = self.submod_94(getitem_236, s59, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_, getitem_237, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_236 = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_ = getitem_237 = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_238 = submod_94[0]
        getitem_239 = submod_94[1]
        getitem_240 = submod_94[2]
        getitem_241 = submod_94[3]
        getitem_242 = submod_94[4];  submod_94 = None
        submod_95 = self.submod_95(getitem_238, s59, getitem_239, getitem_240, getitem_241);  getitem_238 = getitem_239 = getitem_240 = submod_95 = None
        submod_96 = self.submod_96(getitem_241, s59, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_, getitem_242, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_241 = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_ = getitem_242 = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_243 = submod_96[0]
        getitem_244 = submod_96[1]
        getitem_245 = submod_96[2]
        getitem_246 = submod_96[3]
        getitem_247 = submod_96[4];  submod_96 = None
        submod_97 = self.submod_97(getitem_243, s59, getitem_244, getitem_245, getitem_246);  getitem_243 = getitem_244 = getitem_245 = submod_97 = None
        submod_98 = self.submod_98(getitem_246, s59, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_, getitem_247, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_246 = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_ = getitem_247 = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_248 = submod_98[0]
        getitem_249 = submod_98[1]
        getitem_250 = submod_98[2]
        getitem_251 = submod_98[3]
        getitem_252 = submod_98[4];  submod_98 = None
        submod_99 = self.submod_99(getitem_248, s59, getitem_249, getitem_250, getitem_251);  getitem_248 = getitem_249 = getitem_250 = submod_99 = None
        submod_100 = self.submod_100(getitem_251, s59, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_, getitem_252, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_251 = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_ = getitem_252 = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_253 = submod_100[0]
        getitem_254 = submod_100[1]
        getitem_255 = submod_100[2]
        getitem_256 = submod_100[3]
        getitem_257 = submod_100[4];  submod_100 = None
        submod_101 = self.submod_101(getitem_253, s59, getitem_254, getitem_255, getitem_256);  getitem_253 = getitem_254 = getitem_255 = submod_101 = None
        submod_102 = self.submod_102(getitem_256, s59, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_, getitem_257, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_256 = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_ = getitem_257 = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_258 = submod_102[0]
        getitem_259 = submod_102[1]
        getitem_260 = submod_102[2]
        getitem_261 = submod_102[3]
        getitem_262 = submod_102[4];  submod_102 = None
        submod_103 = self.submod_103(getitem_258, s59, getitem_259, getitem_260, getitem_261);  getitem_258 = getitem_259 = getitem_260 = submod_103 = None
        submod_104 = self.submod_104(getitem_261, s59, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_, getitem_262, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_261 = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_ = getitem_262 = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_263 = submod_104[0]
        getitem_264 = submod_104[1]
        getitem_265 = submod_104[2]
        getitem_266 = submod_104[3]
        getitem_267 = submod_104[4];  submod_104 = None
        submod_105 = self.submod_105(getitem_263, s59, getitem_264, getitem_265, getitem_266);  getitem_263 = getitem_264 = getitem_265 = submod_105 = None
        submod_106 = self.submod_106(getitem_266, s59, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_, getitem_267, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_266 = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_ = getitem_267 = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_268 = submod_106[0]
        getitem_269 = submod_106[1]
        getitem_270 = submod_106[2]
        getitem_271 = submod_106[3]
        getitem_272 = submod_106[4];  submod_106 = None
        submod_107 = self.submod_107(getitem_268, s59, getitem_269, getitem_270, getitem_271);  getitem_268 = getitem_269 = getitem_270 = submod_107 = None
        submod_108 = self.submod_108(getitem_271, s59, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_, getitem_272, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_271 = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_ = getitem_272 = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_273 = submod_108[0]
        getitem_274 = submod_108[1]
        getitem_275 = submod_108[2]
        getitem_276 = submod_108[3]
        getitem_277 = submod_108[4];  submod_108 = None
        submod_109 = self.submod_109(getitem_273, s59, getitem_274, getitem_275, getitem_276);  getitem_273 = getitem_274 = getitem_275 = submod_109 = None
        submod_110 = self.submod_110(getitem_276, s59, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_, getitem_277, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_276 = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_ = getitem_277 = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_278 = submod_110[0]
        getitem_279 = submod_110[1]
        getitem_280 = submod_110[2]
        getitem_281 = submod_110[3]
        getitem_282 = submod_110[4];  submod_110 = None
        submod_111 = self.submod_111(getitem_278, s59, getitem_279, getitem_280, getitem_281);  getitem_278 = getitem_279 = getitem_280 = submod_111 = None
        submod_112 = self.submod_112(getitem_281, s59, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_, getitem_282, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_281 = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_ = getitem_282 = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
        getitem_283 = submod_112[0]
        getitem_284 = submod_112[1]
        getitem_285 = submod_112[2]
        getitem_286 = submod_112[3]
        getitem_287 = submod_112[4];  submod_112 = None
        submod_113 = self.submod_113(getitem_283, s59, getitem_284, getitem_285, getitem_286);  getitem_283 = getitem_284 = getitem_285 = submod_113 = None
        submod_114 = self.submod_114(getitem_286, s59, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_, getitem_287, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_, l_positions_, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_);  getitem_286 = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_ = getitem_287 = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_positions_ = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = None
        getitem_288 = submod_114[0]
        getitem_289 = submod_114[1]
        getitem_290 = submod_114[2]
        getitem_291 = submod_114[3]
        getitem_292 = submod_114[4];  submod_114 = None
        submod_115 = self.submod_115(getitem_288, s59, getitem_289, getitem_290, getitem_291);  getitem_288 = getitem_289 = getitem_290 = submod_115 = None
        submod_116 = self.submod_116(getitem_291, s59, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_, getitem_292, getitem_167, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_language_model_modules_model_modules_norm_parameters_weight_);  getitem_291 = s59 = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_ = getitem_292 = getitem_167 = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_language_model_modules_model_modules_norm_parameters_weight_ = None
        return (submod_116,)
        
    class submod_0(torch.nn.Module):
        def forward(self, s59: "Sym(s59)", l_inputs_embeds_: "bf16[s59, 5120]", l_positions_: "i64[s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:360 in forward, code: inputs_embeds.shape[0] * pool_size, inputs_embeds.shape[1] // pool_size
            mul: "Sym(4*s59)" = s59 * 4;  s59 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:359 in forward, code: inputs_embeds = inputs_embeds.view(
            view: "bf16[4*s59, 1280]" = l_inputs_embeds_.view(mul, 1280);  l_inputs_embeds_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:121 in _expand_tensor, code: base = input_tensor * scaling
            mul_1: "i64[s59]" = l_positions_ * 4;  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:124 in _expand_tensor, code: offsets = torch.arange(scaling, device=input_tensor.device)
            arange: "i64[4]" = torch.arange(4, device = device(type='cuda', index=0))
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:128 in _expand_tensor, code: return (base.unsqueeze(1) + offsets).view(-1)
            unsqueeze: "i64[s59, 1]" = mul_1.unsqueeze(1);  mul_1 = None
            add: "i64[s59, 4]" = unsqueeze + arange;  unsqueeze = arange = None
            view_1: "i64[4*s59]" = add.view(-1);  add = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = view.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_2: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_2.to(torch.bfloat16);  mul_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_3: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear.split([2048, 2048, 2048], dim = -1);  linear = None
            getitem: "bf16[4*s59, 2048]" = split[0]
            getitem_1: "bf16[4*s59, 2048]" = split[1]
            getitem_2: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = view_1.flatten()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_3: "bf16[4*s59, 32]" = chunk[0]
            getitem_4: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem.view(mul, -1, 64);  getitem = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_5: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_6: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_3.unsqueeze(-2)
            to_2: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_4.unsqueeze(-2)
            to_3: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_7: "bf16[4*s59, 32, 32]" = getitem_5[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_8: "bf16[4*s59, 32, 32]" = getitem_5[(Ellipsis, slice(1, None, 2))];  getitem_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_4: "bf16[4*s59, 32, 32]" = getitem_7 * to_2
            mul_5: "bf16[4*s59, 32, 32]" = getitem_8 * to_3
            sub: "bf16[4*s59, 32, 32]" = mul_4 - mul_5;  mul_4 = mul_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_8 * to_2;  getitem_8 = to_2 = None
            mul_7: "bf16[4*s59, 32, 32]" = getitem_7 * to_3;  getitem_7 = to_3 = None
            add_2: "bf16[4*s59, 32, 32]" = mul_6 + mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_2), dim = -1);  sub = add_2 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_6), dim = -1);  flatten_1 = getitem_6 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_1.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_3: "bf16[4*s59, 32, 64]" = getitem_1.view(mul, -1, 64);  getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_9: "bf16[4*s59, 32, 64]" = view_3[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_10: "bf16[4*s59, 32, 0]" = view_3[(Ellipsis, slice(64, None, None))];  view_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_3.unsqueeze(-2);  getitem_3 = None
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_4: "bf16[4*s59, 1, 32]" = getitem_4.unsqueeze(-2);  getitem_4 = None
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_4.to(torch.bfloat16);  unsqueeze_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_11: "bf16[4*s59, 32, 32]" = getitem_9[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_12: "bf16[4*s59, 32, 32]" = getitem_9[(Ellipsis, slice(1, None, 2))];  getitem_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_11 * to_4
            mul_9: "bf16[4*s59, 32, 32]" = getitem_12 * to_5
            sub_1: "bf16[4*s59, 32, 32]" = mul_8 - mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_12 * to_4;  getitem_12 = to_4 = None
            mul_11: "bf16[4*s59, 32, 32]" = getitem_11 * to_5;  getitem_11 = to_5 = None
            add_3: "bf16[4*s59, 32, 32]" = mul_10 + mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_3), dim = -1);  sub_1 = add_3 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_10), dim = -1);  flatten_2 = getitem_10 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048])
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_4: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_5: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_6: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_7: "bf16[4*s59, 32, 64]" = getitem_2.view(-1, 32, 64);  getitem_2 = None
            return (view_4, view_6, view_7, view_5, view, view_1, mul)
            
    class submod_1(torch.nn.Module):
        def forward(self, query_2: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_2: "bf16[4*s59, 32, 64]", value: "bf16[4*s59, 32, 64]", output_3: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_2, key_2, value, output_3, 'whisper_encoder.whisper_encoder.layers.0.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_2 = key_2 = value = output_3 = unified_attention_with_output = None
            return ()
            
    class submod_2(torch.nn.Module):
        def forward(self, output_3: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", inputs_embeds: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_3.view(-1, 2048);  output_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = inputs_embeds + linear;  inputs_embeds = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_3(torch.nn.Module):
        def forward(self, query_5: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_5: "bf16[4*s59, 32, 64]", value_1: "bf16[4*s59, 32, 64]", output_7: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_5, key_5, value_1, output_7, 'whisper_encoder.whisper_encoder.layers.1.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_5 = key_5 = value_1 = output_7 = unified_attention_with_output = None
            return ()
            
    class submod_4(torch.nn.Module):
        def forward(self, output_7: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_1: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_7.view(-1, 2048);  output_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_1 + linear;  hidden_states_1 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_5(torch.nn.Module):
        def forward(self, query_8: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_8: "bf16[4*s59, 32, 64]", value_2: "bf16[4*s59, 32, 64]", output_11: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_8, key_8, value_2, output_11, 'whisper_encoder.whisper_encoder.layers.2.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_8 = key_8 = value_2 = output_11 = unified_attention_with_output = None
            return ()
            
    class submod_6(torch.nn.Module):
        def forward(self, output_11: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_3: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_11.view(-1, 2048);  output_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_3 + linear;  hidden_states_3 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_7(torch.nn.Module):
        def forward(self, query_11: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_11: "bf16[4*s59, 32, 64]", value_3: "bf16[4*s59, 32, 64]", output_15: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_11, key_11, value_3, output_15, 'whisper_encoder.whisper_encoder.layers.3.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_11 = key_11 = value_3 = output_15 = unified_attention_with_output = None
            return ()
            
    class submod_8(torch.nn.Module):
        def forward(self, output_15: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_5: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_15.view(-1, 2048);  output_15 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_5 + linear;  hidden_states_5 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_9(torch.nn.Module):
        def forward(self, query_14: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_14: "bf16[4*s59, 32, 64]", value_4: "bf16[4*s59, 32, 64]", output_19: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_14, key_14, value_4, output_19, 'whisper_encoder.whisper_encoder.layers.4.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_14 = key_14 = value_4 = output_19 = unified_attention_with_output = None
            return ()
            
    class submod_10(torch.nn.Module):
        def forward(self, output_19: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_7: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_19.view(-1, 2048);  output_19 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_7 + linear;  hidden_states_7 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_11(torch.nn.Module):
        def forward(self, query_17: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_17: "bf16[4*s59, 32, 64]", value_5: "bf16[4*s59, 32, 64]", output_23: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_17, key_17, value_5, output_23, 'whisper_encoder.whisper_encoder.layers.5.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_17 = key_17 = value_5 = output_23 = unified_attention_with_output = None
            return ()
            
    class submod_12(torch.nn.Module):
        def forward(self, output_23: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_9: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_23.view(-1, 2048);  output_23 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_9 + linear;  hidden_states_9 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_13(torch.nn.Module):
        def forward(self, query_20: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_20: "bf16[4*s59, 32, 64]", value_6: "bf16[4*s59, 32, 64]", output_27: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_20, key_20, value_6, output_27, 'whisper_encoder.whisper_encoder.layers.6.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_20 = key_20 = value_6 = output_27 = unified_attention_with_output = None
            return ()
            
    class submod_14(torch.nn.Module):
        def forward(self, output_27: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_11: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_27.view(-1, 2048);  output_27 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_11 + linear;  hidden_states_11 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_15(torch.nn.Module):
        def forward(self, query_23: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_23: "bf16[4*s59, 32, 64]", value_7: "bf16[4*s59, 32, 64]", output_31: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_23, key_23, value_7, output_31, 'whisper_encoder.whisper_encoder.layers.7.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_23 = key_23 = value_7 = output_31 = unified_attention_with_output = None
            return ()
            
    class submod_16(torch.nn.Module):
        def forward(self, output_31: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_13: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_31.view(-1, 2048);  output_31 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_13 + linear;  hidden_states_13 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_17(torch.nn.Module):
        def forward(self, query_26: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_26: "bf16[4*s59, 32, 64]", value_8: "bf16[4*s59, 32, 64]", output_35: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_26, key_26, value_8, output_35, 'whisper_encoder.whisper_encoder.layers.8.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_26 = key_26 = value_8 = output_35 = unified_attention_with_output = None
            return ()
            
    class submod_18(torch.nn.Module):
        def forward(self, output_35: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_15: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_35.view(-1, 2048);  output_35 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_15 + linear;  hidden_states_15 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_19(torch.nn.Module):
        def forward(self, query_29: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_29: "bf16[4*s59, 32, 64]", value_9: "bf16[4*s59, 32, 64]", output_39: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_29, key_29, value_9, output_39, 'whisper_encoder.whisper_encoder.layers.9.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_29 = key_29 = value_9 = output_39 = unified_attention_with_output = None
            return ()
            
    class submod_20(torch.nn.Module):
        def forward(self, output_39: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_17: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_39.view(-1, 2048);  output_39 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_17 + linear;  hidden_states_17 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_21(torch.nn.Module):
        def forward(self, query_32: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_32: "bf16[4*s59, 32, 64]", value_10: "bf16[4*s59, 32, 64]", output_43: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_32, key_32, value_10, output_43, 'whisper_encoder.whisper_encoder.layers.10.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_32 = key_32 = value_10 = output_43 = unified_attention_with_output = None
            return ()
            
    class submod_22(torch.nn.Module):
        def forward(self, output_43: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_19: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_43.view(-1, 2048);  output_43 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_19 + linear;  hidden_states_19 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_23(torch.nn.Module):
        def forward(self, query_35: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_35: "bf16[4*s59, 32, 64]", value_11: "bf16[4*s59, 32, 64]", output_47: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_35, key_35, value_11, output_47, 'whisper_encoder.whisper_encoder.layers.11.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_35 = key_35 = value_11 = output_47 = unified_attention_with_output = None
            return ()
            
    class submod_24(torch.nn.Module):
        def forward(self, output_47: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_21: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_47.view(-1, 2048);  output_47 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_21 + linear;  hidden_states_21 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_25(torch.nn.Module):
        def forward(self, query_38: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_38: "bf16[4*s59, 32, 64]", value_12: "bf16[4*s59, 32, 64]", output_51: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_38, key_38, value_12, output_51, 'whisper_encoder.whisper_encoder.layers.12.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_38 = key_38 = value_12 = output_51 = unified_attention_with_output = None
            return ()
            
    class submod_26(torch.nn.Module):
        def forward(self, output_51: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_23: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_51.view(-1, 2048);  output_51 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_23 + linear;  hidden_states_23 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_27(torch.nn.Module):
        def forward(self, query_41: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_41: "bf16[4*s59, 32, 64]", value_13: "bf16[4*s59, 32, 64]", output_55: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_41, key_41, value_13, output_55, 'whisper_encoder.whisper_encoder.layers.13.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_41 = key_41 = value_13 = output_55 = unified_attention_with_output = None
            return ()
            
    class submod_28(torch.nn.Module):
        def forward(self, output_55: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_25: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_55.view(-1, 2048);  output_55 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_25 + linear;  hidden_states_25 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_29(torch.nn.Module):
        def forward(self, query_44: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_44: "bf16[4*s59, 32, 64]", value_14: "bf16[4*s59, 32, 64]", output_59: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_44, key_44, value_14, output_59, 'whisper_encoder.whisper_encoder.layers.14.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_44 = key_44 = value_14 = output_59 = unified_attention_with_output = None
            return ()
            
    class submod_30(torch.nn.Module):
        def forward(self, output_59: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_27: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_59.view(-1, 2048);  output_59 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_27 + linear;  hidden_states_27 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_31(torch.nn.Module):
        def forward(self, query_47: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_47: "bf16[4*s59, 32, 64]", value_15: "bf16[4*s59, 32, 64]", output_63: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_47, key_47, value_15, output_63, 'whisper_encoder.whisper_encoder.layers.15.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_47 = key_47 = value_15 = output_63 = unified_attention_with_output = None
            return ()
            
    class submod_32(torch.nn.Module):
        def forward(self, output_63: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_29: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_63.view(-1, 2048);  output_63 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_29 + linear;  hidden_states_29 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_33(torch.nn.Module):
        def forward(self, query_50: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_50: "bf16[4*s59, 32, 64]", value_16: "bf16[4*s59, 32, 64]", output_67: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_50, key_50, value_16, output_67, 'whisper_encoder.whisper_encoder.layers.16.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_50 = key_50 = value_16 = output_67 = unified_attention_with_output = None
            return ()
            
    class submod_34(torch.nn.Module):
        def forward(self, output_67: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_31: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_67.view(-1, 2048);  output_67 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_31 + linear;  hidden_states_31 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_35(torch.nn.Module):
        def forward(self, query_53: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_53: "bf16[4*s59, 32, 64]", value_17: "bf16[4*s59, 32, 64]", output_71: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_53, key_53, value_17, output_71, 'whisper_encoder.whisper_encoder.layers.17.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_53 = key_53 = value_17 = output_71 = unified_attention_with_output = None
            return ()
            
    class submod_36(torch.nn.Module):
        def forward(self, output_71: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_33: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_71.view(-1, 2048);  output_71 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_33 + linear;  hidden_states_33 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_37(torch.nn.Module):
        def forward(self, query_56: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_56: "bf16[4*s59, 32, 64]", value_18: "bf16[4*s59, 32, 64]", output_75: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_56, key_56, value_18, output_75, 'whisper_encoder.whisper_encoder.layers.18.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_56 = key_56 = value_18 = output_75 = unified_attention_with_output = None
            return ()
            
    class submod_38(torch.nn.Module):
        def forward(self, output_75: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_35: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_75.view(-1, 2048);  output_75 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_35 + linear;  hidden_states_35 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_39(torch.nn.Module):
        def forward(self, query_59: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_59: "bf16[4*s59, 32, 64]", value_19: "bf16[4*s59, 32, 64]", output_79: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_59, key_59, value_19, output_79, 'whisper_encoder.whisper_encoder.layers.19.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_59 = key_59 = value_19 = output_79 = unified_attention_with_output = None
            return ()
            
    class submod_40(torch.nn.Module):
        def forward(self, output_79: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_37: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_79.view(-1, 2048);  output_79 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_37 + linear;  hidden_states_37 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_41(torch.nn.Module):
        def forward(self, query_62: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_62: "bf16[4*s59, 32, 64]", value_20: "bf16[4*s59, 32, 64]", output_83: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_62, key_62, value_20, output_83, 'whisper_encoder.whisper_encoder.layers.20.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_62 = key_62 = value_20 = output_83 = unified_attention_with_output = None
            return ()
            
    class submod_42(torch.nn.Module):
        def forward(self, output_83: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_39: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_83.view(-1, 2048);  output_83 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_39 + linear;  hidden_states_39 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_43(torch.nn.Module):
        def forward(self, query_65: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_65: "bf16[4*s59, 32, 64]", value_21: "bf16[4*s59, 32, 64]", output_87: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_65, key_65, value_21, output_87, 'whisper_encoder.whisper_encoder.layers.21.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_65 = key_65 = value_21 = output_87 = unified_attention_with_output = None
            return ()
            
    class submod_44(torch.nn.Module):
        def forward(self, output_87: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_41: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_87.view(-1, 2048);  output_87 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_41 + linear;  hidden_states_41 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_45(torch.nn.Module):
        def forward(self, query_68: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_68: "bf16[4*s59, 32, 64]", value_22: "bf16[4*s59, 32, 64]", output_91: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_68, key_68, value_22, output_91, 'whisper_encoder.whisper_encoder.layers.22.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_68 = key_68 = value_22 = output_91 = unified_attention_with_output = None
            return ()
            
    class submod_46(torch.nn.Module):
        def forward(self, output_91: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_43: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_91.view(-1, 2048);  output_91 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_43 + linear;  hidden_states_43 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_47(torch.nn.Module):
        def forward(self, query_71: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_71: "bf16[4*s59, 32, 64]", value_23: "bf16[4*s59, 32, 64]", output_95: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_71, key_71, value_23, output_95, 'whisper_encoder.whisper_encoder.layers.23.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_71 = key_71 = value_23 = output_95 = unified_attention_with_output = None
            return ()
            
    class submod_48(torch.nn.Module):
        def forward(self, output_95: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_45: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_95.view(-1, 2048);  output_95 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_45 + linear;  hidden_states_45 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_49(torch.nn.Module):
        def forward(self, query_74: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_74: "bf16[4*s59, 32, 64]", value_24: "bf16[4*s59, 32, 64]", output_99: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_74, key_74, value_24, output_99, 'whisper_encoder.whisper_encoder.layers.24.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_74 = key_74 = value_24 = output_99 = unified_attention_with_output = None
            return ()
            
    class submod_50(torch.nn.Module):
        def forward(self, output_99: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_47: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_99.view(-1, 2048);  output_99 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_47 + linear;  hidden_states_47 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_51(torch.nn.Module):
        def forward(self, query_77: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_77: "bf16[4*s59, 32, 64]", value_25: "bf16[4*s59, 32, 64]", output_103: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_77, key_77, value_25, output_103, 'whisper_encoder.whisper_encoder.layers.25.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_77 = key_77 = value_25 = output_103 = unified_attention_with_output = None
            return ()
            
    class submod_52(torch.nn.Module):
        def forward(self, output_103: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_49: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_103.view(-1, 2048);  output_103 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_49 + linear;  hidden_states_49 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_53(torch.nn.Module):
        def forward(self, query_80: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_80: "bf16[4*s59, 32, 64]", value_26: "bf16[4*s59, 32, 64]", output_107: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_80, key_80, value_26, output_107, 'whisper_encoder.whisper_encoder.layers.26.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_80 = key_80 = value_26 = output_107 = unified_attention_with_output = None
            return ()
            
    class submod_54(torch.nn.Module):
        def forward(self, output_107: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_51: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_107.view(-1, 2048);  output_107 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_51 + linear;  hidden_states_51 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_26_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_55(torch.nn.Module):
        def forward(self, query_83: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_83: "bf16[4*s59, 32, 64]", value_27: "bf16[4*s59, 32, 64]", output_111: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_83, key_83, value_27, output_111, 'whisper_encoder.whisper_encoder.layers.27.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_83 = key_83 = value_27 = output_111 = unified_attention_with_output = None
            return ()
            
    class submod_56(torch.nn.Module):
        def forward(self, output_111: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_53: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_111.view(-1, 2048);  output_111 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_53 + linear;  hidden_states_53 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_27_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_57(torch.nn.Module):
        def forward(self, query_86: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_86: "bf16[4*s59, 32, 64]", value_28: "bf16[4*s59, 32, 64]", output_115: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_86, key_86, value_28, output_115, 'whisper_encoder.whisper_encoder.layers.28.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_86 = key_86 = value_28 = output_115 = unified_attention_with_output = None
            return ()
            
    class submod_58(torch.nn.Module):
        def forward(self, output_115: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_55: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_115.view(-1, 2048);  output_115 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_55 + linear;  hidden_states_55 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_28_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_59(torch.nn.Module):
        def forward(self, query_89: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_89: "bf16[4*s59, 32, 64]", value_29: "bf16[4*s59, 32, 64]", output_119: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_89, key_89, value_29, output_119, 'whisper_encoder.whisper_encoder.layers.29.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_89 = key_89 = value_29 = output_119 = unified_attention_with_output = None
            return ()
            
    class submod_60(torch.nn.Module):
        def forward(self, output_119: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_57: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_119.view(-1, 2048);  output_119 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_57 + linear;  hidden_states_57 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_29_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_61(torch.nn.Module):
        def forward(self, query_92: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_92: "bf16[4*s59, 32, 64]", value_30: "bf16[4*s59, 32, 64]", output_123: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_92, key_92, value_30, output_123, 'whisper_encoder.whisper_encoder.layers.30.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_92 = key_92 = value_30 = output_123 = unified_attention_with_output = None
            return ()
            
    class submod_62(torch.nn.Module):
        def forward(self, output_123: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_59: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_: "bf16[6144]", whisper_positions: "i64[4*s59]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[524288, 64]", mul: "Sym(4*s59)"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_123.view(-1, 2048);  output_123 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_59 + linear;  hidden_states_59 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul_1: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul_1.to(torch.bfloat16);  mul_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_2: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_3: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_3, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_);  mul_3 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_30_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_4: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[4*s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_);  mul_5 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_qkv_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:433 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_3.split([2048, 2048, 2048], dim = -1);  linear_3 = None
            getitem_2: "bf16[4*s59, 2048]" = split[0]
            getitem_3: "bf16[4*s59, 2048]" = split[1]
            getitem_4: "bf16[4*s59, 2048]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[4*s59]" = whisper_positions.flatten();  whisper_positions = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[4*s59, 64]" = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[4*s59, 32]" = chunk[0]
            getitem_6: "bf16[4*s59, 32]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[4*s59, 32, 64]" = getitem_2.view(mul, -1, 64);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[4*s59, 32, 64]" = view_1[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[4*s59, 32, 0]" = view_1[(Ellipsis, slice(64, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2)
            to_4: "bf16[4*s59, 1, 32]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2)
            to_5: "bf16[4*s59, 1, 32]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_9: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_10: "bf16[4*s59, 32, 32]" = getitem_7[(Ellipsis, slice(1, None, 2))];  getitem_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[4*s59, 32, 32]" = getitem_9 * to_4
            mul_7: "bf16[4*s59, 32, 32]" = getitem_10 * to_5
            sub: "bf16[4*s59, 32, 32]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[4*s59, 32, 32]" = getitem_10 * to_4;  getitem_10 = to_4 = None
            mul_9: "bf16[4*s59, 32, 32]" = getitem_9 * to_5;  getitem_9 = to_5 = None
            add_4: "bf16[4*s59, 32, 32]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub, add_4), dim = -1);  sub = add_4 = None
            flatten_1: "bf16[4*s59, 32, 64]" = stack.flatten(-2);  stack = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat: "bf16[4*s59, 32, 64]" = torch.cat((flatten_1, getitem_8), dim = -1);  flatten_1 = getitem_8 = None
            reshape: "bf16[4*s59, 2048]" = cat.reshape(size);  cat = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[4*s59, 32, 64]" = getitem_3.view(mul, -1, 64);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[4*s59, 32, 64]" = view_2[(Ellipsis, slice(None, 64, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[4*s59, 32, 0]" = view_2[(Ellipsis, slice(64, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[4*s59, 1, 32]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_6: "bf16[4*s59, 1, 32]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[4*s59, 1, 32]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_7: "bf16[4*s59, 1, 32]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:169 in forward_static, code: x1 = x[..., ::2]
            getitem_13: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(None, None, 2))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:170 in forward_static, code: x2 = x[..., 1::2]
            getitem_14: "bf16[4*s59, 32, 32]" = getitem_11[(Ellipsis, slice(1, None, 2))];  getitem_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[4*s59, 32, 32]" = getitem_13 * to_6
            mul_11: "bf16[4*s59, 32, 32]" = getitem_14 * to_7
            sub_1: "bf16[4*s59, 32, 32]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[4*s59, 32, 32]" = getitem_14 * to_6;  getitem_14 = to_6 = None
            mul_13: "bf16[4*s59, 32, 32]" = getitem_13 * to_7;  getitem_13 = to_7 = None
            add_5: "bf16[4*s59, 32, 32]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:178 in forward_static, code: output = torch.stack((o1, o2), dim=-1).flatten(-2)
            stack_1: "bf16[4*s59, 32, 32, 2]" = torch.stack((sub_1, add_5), dim = -1);  sub_1 = add_5 = None
            flatten_2: "bf16[4*s59, 32, 64]" = stack_1.flatten(-2);  stack_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_1: "bf16[4*s59, 32, 64]" = torch.cat((flatten_2, getitem_12), dim = -1);  flatten_2 = getitem_12 = None
            reshape_1: "bf16[4*s59, 2048]" = cat_1.reshape(size_1);  cat_1 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([mul, 2048]);  mul = None
            empty: "bf16[4*s59, 2048]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[4*s59, 32, 64]" = reshape.view(-1, 32, 64);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[4*s59, 32, 64]" = empty.view(-1, 32, 64);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[4*s59, 32, 64]" = reshape_1.view(-1, 32, 64);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[4*s59, 32, 64]" = getitem_4.view(-1, 32, 64);  getitem_4 = None
            return (view_3, view_5, view_6, view_4, add_2)
            
    class submod_63(torch.nn.Module):
        def forward(self, query_95: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", key_95: "bf16[4*s59, 32, 64]", value_31: "bf16[4*s59, 32, 64]", output_127: "bf16[4*s59, 32, 64]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_95, key_95, value_31, output_127, 'whisper_encoder.whisper_encoder.layers.31.layers.self_attn.attn', kv_cache_dummy_dep = None);  query_95 = key_95 = value_31 = output_127 = unified_attention_with_output = None
            return ()
            
    class submod_64(torch.nn.Module):
        def forward(self, output_127: "bf16[4*s59, 32, 64]", s59: "Sym(s59)", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_: "bf16[1280, 2048]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_: "bf16[1280]", hidden_states_61: "bf16[4*s59, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[10240, 1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_: "bf16[1280, 5120]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_: "bf16[1280]", l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_: "bf16[1280]", l_self_modules_audio_language_adapter_modules_w_in_parameters_weight_: "bf16[3072, 5120]", l_self_modules_audio_language_adapter_modules_w_out_parameters_weight_: "bf16[3072, 3072]", l_input_ids_: "i32[s59]", l_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_: "bf16[131072, 3072]", l_self_modules_time_embedding_buffers_inv_freq_: "f32[1536]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[4*s59, 2048]" = output_127.view(-1, 2048);  output_127 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[4*s59, 1280]" = torch._C._nn.linear(view, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_);  view = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_self_attn_modules_out_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:490 in forward, code: hidden_states = residual + hidden_states
            add: "bf16[4*s59, 1280]" = hidden_states_61 + linear;  hidden_states_61 = linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_final_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[4*s59, 1280]" = add.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[4*s59, 1280]" = to.pow(2)
            mean: "f32[4*s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[4*s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[4*s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[4*s59, 1280]" = to * rsqrt;  to = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_1: "bf16[4*s59, 1280]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[4*s59, 1280]" = to_1 * _get_data_attr;  to_1 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[4*s59, 10240]" = torch._C._nn.linear(mul_1, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_1 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(None, 5120, None))]
            silu: "bf16[4*s59, 5120]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[4*s59, 5120]" = linear_1[(Ellipsis, slice(5120, None, None))];  linear_1 = None
            mul_2: "bf16[4*s59, 5120]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[4*s59, 1280]" = torch._C._nn.linear(mul_2, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_, l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_);  mul_2 = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_weight_ = l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layers_modules_31_modules_mlp_modules_down_proj_parameters_bias_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/whisper_causal.py:494 in forward, code: hidden_states = residual + hidden_states
            add_2: "bf16[4*s59, 1280]" = add + linear_2;  add = linear_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[1280]" = torch._C._autograd._get_data_attr(l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_);  l_self_modules_whisper_encoder_modules_whisper_encoder_modules_layer_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_2: "f32[4*s59, 1280]" = add_2.to(torch.float32);  add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[4*s59, 1280]" = to_2.pow(2)
            mean_1: "f32[4*s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_3: "f32[4*s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[4*s59, 1]" = torch.rsqrt(add_3);  add_3 = None
            mul_3: "f32[4*s59, 1280]" = to_2 * rsqrt_1;  to_2 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_3: "bf16[4*s59, 1280]" = mul_3.to(torch.bfloat16);  mul_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_4: "bf16[4*s59, 1280]" = to_3 * _get_data_attr_1;  to_3 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:370 in forward, code: audio_hidden_states = audio_hidden_states.reshape(
            reshape: "bf16[s59, 5120]" = mul_4.reshape(s59, 5120);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral.py:742 in forward, code: return self.w_out(self.gelu(self.w_in(x)))
            linear_3: "bf16[s59, 3072]" = torch._C._nn.linear(reshape, l_self_modules_audio_language_adapter_modules_w_in_parameters_weight_, None);  reshape = l_self_modules_audio_language_adapter_modules_w_in_parameters_weight_ = None
            gelu: "bf16[s59, 3072]" = torch._C._nn.gelu(linear_3, approximate = 'none');  linear_3 = None
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(gelu, l_self_modules_audio_language_adapter_modules_w_out_parameters_weight_, None);  gelu = l_self_modules_audio_language_adapter_modules_w_out_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/vocab_parallel_embedding.py:478 in forward_native, code: output_parallel = self.quant_method.embedding(self, masked_input.long())
            long: "i64[s59]" = l_input_ids_.long();  l_input_ids_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/vocab_parallel_embedding.py:72 in embedding, code: return F.embedding(input_, layer.weight)
            embedding: "bf16[s59, 3072]" = torch.nn.functional.embedding(long, l_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_);  long = l_self_modules_language_model_modules_model_modules_embed_tokens_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:379 in forward, code: inputs_embeds = audio_text_embeds + text_embeds
            add_4: "bf16[s59, 3072]" = linear_4 + embedding;  linear_4 = embedding = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:381 in forward, code: time_tensor = torch.full(
            full: "bf16[1]" = torch.full((1,), fill_value = 3, device = device(type='cuda', index=0), dtype = torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:111 in forward, code: t = t[..., None]  # (B,) -> (B, 1) or (B, T) -> (B, T, 1)
            getitem_2: "bf16[1, 1]" = full[(Ellipsis, None)];  full = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:112 in forward, code: inv_freq = self.inv_freq.to(device=t.device, dtype=t.dtype)
            to_4: "bf16[1536]" = l_self_modules_time_embedding_buffers_inv_freq_.to(device = device(type='cuda', index=0), dtype = torch.bfloat16);  l_self_modules_time_embedding_buffers_inv_freq_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:114 in forward, code: t * inv_freq
            mul_5: "bf16[1, 1536]" = getitem_2 * to_4;  getitem_2 = to_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/voxtral_realtime.py:116 in forward, code: return torch.cat((emb.cos(), emb.sin()), dim=-1)  # (B, D) or (B, T, D)
            cos: "bf16[1, 1536]" = mul_5.cos()
            sin: "bf16[1, 1536]" = mul_5.sin();  mul_5 = None
            cat: "bf16[1, 3072]" = torch.cat((cos, sin), dim = -1);  cos = sin = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_2: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_5: "f32[s59, 3072]" = add_4.to(torch.float32)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_3: "f32[s59, 3072]" = to_5.pow(2)
            mean_2: "f32[s59, 1]" = pow_3.mean(dim = -1, keepdim = True);  pow_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_5: "f32[s59, 1]" = mean_2 + 1e-05;  mean_2 = None
            rsqrt_2: "f32[s59, 1]" = torch.rsqrt(add_5);  add_5 = None
            mul_6: "f32[s59, 3072]" = to_5 * rsqrt_2;  to_5 = rsqrt_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_6: "bf16[s59, 3072]" = mul_6.to(torch.bfloat16);  mul_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_7: "bf16[s59, 3072]" = to_6 * _get_data_attr_2;  to_6 = _get_data_attr_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_7, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_7 = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_3: "bf16[s59, 4096]" = split[0]
            getitem_4: "bf16[s59, 1024]" = split[1]
            getitem_5: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_6: "bf16[s59, 64]" = chunk[0]
            getitem_7: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_8: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_9: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_7.unsqueeze(-2)
            to_8: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_8, 2, dim = -1);  getitem_8 = None
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_11: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_7
            mul_9: "bf16[s59, 32, 64]" = getitem_11 * to_8
            sub: "bf16[s59, 32, 64]" = mul_8 - mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_10: "bf16[s59, 32, 64]" = getitem_11 * to_7;  getitem_11 = to_7 = None
            mul_11: "bf16[s59, 32, 64]" = getitem_10 * to_8;  getitem_10 = to_8 = None
            add_6: "bf16[s59, 32, 64]" = mul_10 + mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((sub, add_6), dim = -1);  sub = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_2: "bf16[s59, 32, 128]" = torch.cat((cat_1, getitem_9), dim = -1);  cat_1 = getitem_9 = None
            reshape_1: "bf16[s59, 4096]" = cat_2.reshape(size);  cat_2 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_4.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_4.view(s59, -1, 128);  getitem_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_12: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_13: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_7.unsqueeze(-2);  getitem_7 = None
            to_10: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_12, 2, dim = -1);  getitem_12 = None
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_15: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_9
            mul_13: "bf16[s59, 8, 64]" = getitem_15 * to_10
            sub_1: "bf16[s59, 8, 64]" = mul_12 - mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_14: "bf16[s59, 8, 64]" = getitem_15 * to_9;  getitem_15 = to_9 = None
            mul_15: "bf16[s59, 8, 64]" = getitem_14 * to_10;  getitem_14 = to_10 = None
            add_7: "bf16[s59, 8, 64]" = mul_14 + mul_15;  mul_14 = mul_15 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_7), dim = -1);  sub_1 = add_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_4: "bf16[s59, 8, 128]" = torch.cat((cat_3, getitem_13), dim = -1);  cat_3 = getitem_13 = None
            reshape_2: "bf16[s59, 1024]" = cat_4.reshape(size_1);  cat_4 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape_1.view(-1, 32, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_2.view(-1, 8, 128);  reshape_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_5.view(-1, 8, 128);  getitem_5 = None
            return (view_5, view_6, view_3, view_4, add_4, cat)
            
    class submod_65(torch.nn.Module):
        def forward(self, key_98: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_32: "bf16[s59, 8, 128]", query_98: "bf16[s59, 32, 128]", output_131: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_98, value_32, 'language_model.model.layers.0.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_98, key_98, value_32, output_131, 'language_model.model.layers.0.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_98 = key_98 = value_32 = output_131 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_66(torch.nn.Module):
        def forward(self, output_131: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", inputs_embeds_1: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_131.view(-1, 4096);  output_131 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + inputs_embeds_1;  to = inputs_embeds_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_67(torch.nn.Module):
        def forward(self, key_101: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_33: "bf16[s59, 8, 128]", query_101: "bf16[s59, 32, 128]", output_135: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_101, value_33, 'language_model.model.layers.1.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_101, key_101, value_33, output_135, 'language_model.model.layers.1.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_101 = key_101 = value_33 = output_135 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_68(torch.nn.Module):
        def forward(self, output_135: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_1: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_135.view(-1, 4096);  output_135 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_1;  to = residual_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_1_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_69(torch.nn.Module):
        def forward(self, key_104: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_34: "bf16[s59, 8, 128]", query_104: "bf16[s59, 32, 128]", output_139: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_104, value_34, 'language_model.model.layers.2.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_104, key_104, value_34, output_139, 'language_model.model.layers.2.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_104 = key_104 = value_34 = output_139 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_70(torch.nn.Module):
        def forward(self, output_139: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_3: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_139.view(-1, 4096);  output_139 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_3;  to = residual_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_2_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_71(torch.nn.Module):
        def forward(self, key_107: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_35: "bf16[s59, 8, 128]", query_107: "bf16[s59, 32, 128]", output_143: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_107, value_35, 'language_model.model.layers.3.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_107, key_107, value_35, output_143, 'language_model.model.layers.3.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_107 = key_107 = value_35 = output_143 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_72(torch.nn.Module):
        def forward(self, output_143: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_5: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_143.view(-1, 4096);  output_143 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_5;  to = residual_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_3_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_73(torch.nn.Module):
        def forward(self, key_110: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_36: "bf16[s59, 8, 128]", query_110: "bf16[s59, 32, 128]", output_147: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_110, value_36, 'language_model.model.layers.4.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_110, key_110, value_36, output_147, 'language_model.model.layers.4.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_110 = key_110 = value_36 = output_147 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_74(torch.nn.Module):
        def forward(self, output_147: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_7: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_147.view(-1, 4096);  output_147 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_7;  to = residual_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_4_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_75(torch.nn.Module):
        def forward(self, key_113: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_37: "bf16[s59, 8, 128]", query_113: "bf16[s59, 32, 128]", output_151: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_113, value_37, 'language_model.model.layers.5.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_113, key_113, value_37, output_151, 'language_model.model.layers.5.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_113 = key_113 = value_37 = output_151 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_76(torch.nn.Module):
        def forward(self, output_151: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_9: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_151.view(-1, 4096);  output_151 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_9;  to = residual_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_5_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_77(torch.nn.Module):
        def forward(self, key_116: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_38: "bf16[s59, 8, 128]", query_116: "bf16[s59, 32, 128]", output_155: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_116, value_38, 'language_model.model.layers.6.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_116, key_116, value_38, output_155, 'language_model.model.layers.6.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_116 = key_116 = value_38 = output_155 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_78(torch.nn.Module):
        def forward(self, output_155: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_11: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_155.view(-1, 4096);  output_155 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_11;  to = residual_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_6_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_79(torch.nn.Module):
        def forward(self, key_119: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_39: "bf16[s59, 8, 128]", query_119: "bf16[s59, 32, 128]", output_159: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_119, value_39, 'language_model.model.layers.7.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_119, key_119, value_39, output_159, 'language_model.model.layers.7.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_119 = key_119 = value_39 = output_159 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_80(torch.nn.Module):
        def forward(self, output_159: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_13: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_159.view(-1, 4096);  output_159 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_13;  to = residual_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_7_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_81(torch.nn.Module):
        def forward(self, key_122: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_40: "bf16[s59, 8, 128]", query_122: "bf16[s59, 32, 128]", output_163: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_122, value_40, 'language_model.model.layers.8.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_122, key_122, value_40, output_163, 'language_model.model.layers.8.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_122 = key_122 = value_40 = output_163 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_82(torch.nn.Module):
        def forward(self, output_163: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_15: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_163.view(-1, 4096);  output_163 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_15;  to = residual_15 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_8_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_83(torch.nn.Module):
        def forward(self, key_125: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_41: "bf16[s59, 8, 128]", query_125: "bf16[s59, 32, 128]", output_167: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_125, value_41, 'language_model.model.layers.9.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_125, key_125, value_41, output_167, 'language_model.model.layers.9.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_125 = key_125 = value_41 = output_167 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_84(torch.nn.Module):
        def forward(self, output_167: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_17: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_167.view(-1, 4096);  output_167 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_17;  to = residual_17 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_9_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_85(torch.nn.Module):
        def forward(self, key_128: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_42: "bf16[s59, 8, 128]", query_128: "bf16[s59, 32, 128]", output_171: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_128, value_42, 'language_model.model.layers.10.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_128, key_128, value_42, output_171, 'language_model.model.layers.10.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_128 = key_128 = value_42 = output_171 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_86(torch.nn.Module):
        def forward(self, output_171: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_19: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_171.view(-1, 4096);  output_171 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_19;  to = residual_19 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_10_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_87(torch.nn.Module):
        def forward(self, key_131: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_43: "bf16[s59, 8, 128]", query_131: "bf16[s59, 32, 128]", output_175: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_131, value_43, 'language_model.model.layers.11.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_131, key_131, value_43, output_175, 'language_model.model.layers.11.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_131 = key_131 = value_43 = output_175 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_88(torch.nn.Module):
        def forward(self, output_175: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_21: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_175.view(-1, 4096);  output_175 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_21;  to = residual_21 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_11_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_89(torch.nn.Module):
        def forward(self, key_134: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_44: "bf16[s59, 8, 128]", query_134: "bf16[s59, 32, 128]", output_179: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_134, value_44, 'language_model.model.layers.12.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_134, key_134, value_44, output_179, 'language_model.model.layers.12.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_134 = key_134 = value_44 = output_179 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_90(torch.nn.Module):
        def forward(self, output_179: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_23: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_179.view(-1, 4096);  output_179 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_23;  to = residual_23 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_12_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_91(torch.nn.Module):
        def forward(self, key_137: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_45: "bf16[s59, 8, 128]", query_137: "bf16[s59, 32, 128]", output_183: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_137, value_45, 'language_model.model.layers.13.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_137, key_137, value_45, output_183, 'language_model.model.layers.13.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_137 = key_137 = value_45 = output_183 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_92(torch.nn.Module):
        def forward(self, output_183: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_25: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_183.view(-1, 4096);  output_183 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_25;  to = residual_25 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_13_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_93(torch.nn.Module):
        def forward(self, key_140: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_46: "bf16[s59, 8, 128]", query_140: "bf16[s59, 32, 128]", output_187: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_140, value_46, 'language_model.model.layers.14.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_140, key_140, value_46, output_187, 'language_model.model.layers.14.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_140 = key_140 = value_46 = output_187 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_94(torch.nn.Module):
        def forward(self, output_187: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_27: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_187.view(-1, 4096);  output_187 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_27;  to = residual_27 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_14_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_95(torch.nn.Module):
        def forward(self, key_143: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_47: "bf16[s59, 8, 128]", query_143: "bf16[s59, 32, 128]", output_191: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_143, value_47, 'language_model.model.layers.15.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_143, key_143, value_47, output_191, 'language_model.model.layers.15.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_143 = key_143 = value_47 = output_191 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_96(torch.nn.Module):
        def forward(self, output_191: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_29: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_191.view(-1, 4096);  output_191 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_29;  to = residual_29 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_15_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_97(torch.nn.Module):
        def forward(self, key_146: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_48: "bf16[s59, 8, 128]", query_146: "bf16[s59, 32, 128]", output_195: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_146, value_48, 'language_model.model.layers.16.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_146, key_146, value_48, output_195, 'language_model.model.layers.16.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_146 = key_146 = value_48 = output_195 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_98(torch.nn.Module):
        def forward(self, output_195: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_31: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_195.view(-1, 4096);  output_195 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_31;  to = residual_31 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_16_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_99(torch.nn.Module):
        def forward(self, key_149: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_49: "bf16[s59, 8, 128]", query_149: "bf16[s59, 32, 128]", output_199: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_149, value_49, 'language_model.model.layers.17.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_149, key_149, value_49, output_199, 'language_model.model.layers.17.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_149 = key_149 = value_49 = output_199 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_100(torch.nn.Module):
        def forward(self, output_199: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_33: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_199.view(-1, 4096);  output_199 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_33;  to = residual_33 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_17_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_101(torch.nn.Module):
        def forward(self, key_152: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_50: "bf16[s59, 8, 128]", query_152: "bf16[s59, 32, 128]", output_203: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_152, value_50, 'language_model.model.layers.18.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_152, key_152, value_50, output_203, 'language_model.model.layers.18.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_152 = key_152 = value_50 = output_203 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_102(torch.nn.Module):
        def forward(self, output_203: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_35: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_203.view(-1, 4096);  output_203 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_35;  to = residual_35 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_18_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_103(torch.nn.Module):
        def forward(self, key_155: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_51: "bf16[s59, 8, 128]", query_155: "bf16[s59, 32, 128]", output_207: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_155, value_51, 'language_model.model.layers.19.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_155, key_155, value_51, output_207, 'language_model.model.layers.19.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_155 = key_155 = value_51 = output_207 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_104(torch.nn.Module):
        def forward(self, output_207: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_37: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_207.view(-1, 4096);  output_207 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_37;  to = residual_37 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_19_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_105(torch.nn.Module):
        def forward(self, key_158: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_52: "bf16[s59, 8, 128]", query_158: "bf16[s59, 32, 128]", output_211: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_158, value_52, 'language_model.model.layers.20.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_158, key_158, value_52, output_211, 'language_model.model.layers.20.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_158 = key_158 = value_52 = output_211 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_106(torch.nn.Module):
        def forward(self, output_211: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_39: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_211.view(-1, 4096);  output_211 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_39;  to = residual_39 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_20_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_107(torch.nn.Module):
        def forward(self, key_161: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_53: "bf16[s59, 8, 128]", query_161: "bf16[s59, 32, 128]", output_215: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_161, value_53, 'language_model.model.layers.21.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_161, key_161, value_53, output_215, 'language_model.model.layers.21.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_161 = key_161 = value_53 = output_215 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_108(torch.nn.Module):
        def forward(self, output_215: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_41: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_215.view(-1, 4096);  output_215 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_41;  to = residual_41 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_21_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_109(torch.nn.Module):
        def forward(self, key_164: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_54: "bf16[s59, 8, 128]", query_164: "bf16[s59, 32, 128]", output_219: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_164, value_54, 'language_model.model.layers.22.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_164, key_164, value_54, output_219, 'language_model.model.layers.22.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_164 = key_164 = value_54 = output_219 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_110(torch.nn.Module):
        def forward(self, output_219: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_43: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_219.view(-1, 4096);  output_219 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_43;  to = residual_43 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_22_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_111(torch.nn.Module):
        def forward(self, key_167: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_55: "bf16[s59, 8, 128]", query_167: "bf16[s59, 32, 128]", output_223: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_167, value_55, 'language_model.model.layers.23.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_167, key_167, value_55, output_223, 'language_model.model.layers.23.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_167 = key_167 = value_55 = output_223 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_112(torch.nn.Module):
        def forward(self, output_223: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_45: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_223.view(-1, 4096);  output_223 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_45;  to = residual_45 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_23_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_113(torch.nn.Module):
        def forward(self, key_170: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_56: "bf16[s59, 8, 128]", query_170: "bf16[s59, 32, 128]", output_227: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_170, value_56, 'language_model.model.layers.24.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_170, key_170, value_56, output_227, 'language_model.model.layers.24.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_170 = key_170 = value_56 = output_227 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_114(torch.nn.Module):
        def forward(self, output_227: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_47: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_: "bf16[3072]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_: "bf16[6144, 3072]", l_positions_: "i64[s59]", l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_: "bf16[131072, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_227.view(-1, 4096);  output_227 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_47;  to = residual_47 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_24_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_input_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_5: "bf16[s59, 6144]" = torch._C._nn.linear(mul_5, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_, None);  mul_5 = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_qkv_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:134 in forward, code: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
            split = linear_5.split([4096, 1024, 1024], dim = -1);  linear_5 = None
            getitem_2: "bf16[s59, 4096]" = split[0]
            getitem_3: "bf16[s59, 1024]" = split[1]
            getitem_4: "bf16[s59, 1024]" = split[2];  split = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:145 in forward_static, code: positions = positions.flatten()
            flatten: "i64[s59]" = l_positions_.flatten();  l_positions_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:147 in forward_static, code: cos_sin = cos_sin_cache.index_select(0, positions)
            index_select: "bf16[s59, 128]" = l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_.index_select(0, flatten);  l_self_modules_language_model_modules_model_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache_ = flatten = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:148 in forward_static, code: cos, sin = cos_sin.chunk(2, dim=-1)
            chunk = index_select.chunk(2, dim = -1);  index_select = None
            getitem_5: "bf16[s59, 64]" = chunk[0]
            getitem_6: "bf16[s59, 64]" = chunk[1];  chunk = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:150 in forward_static, code: query_shape = query.shape
            size = getitem_2.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:151 in forward_static, code: query = query.view(num_tokens, -1, head_size)
            view_1: "bf16[s59, 32, 128]" = getitem_2.view(s59, -1, 128);  getitem_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:152 in forward_static, code: query_rot = query[..., :rotary_dim]
            getitem_7: "bf16[s59, 32, 128]" = view_1[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:153 in forward_static, code: query_pass = query[..., rotary_dim:]
            getitem_8: "bf16[s59, 32, 0]" = view_1[(Ellipsis, slice(128, None, None))];  view_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2)
            to_6: "bf16[s59, 1, 64]" = unsqueeze.to(torch.bfloat16);  unsqueeze = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_1: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2)
            to_7: "bf16[s59, 1, 64]" = unsqueeze_1.to(torch.bfloat16);  unsqueeze_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_1 = torch.chunk(getitem_7, 2, dim = -1);  getitem_7 = None
            getitem_9: "bf16[s59, 32, 64]" = chunk_1[0]
            getitem_10: "bf16[s59, 32, 64]" = chunk_1[1];  chunk_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_6: "bf16[s59, 32, 64]" = getitem_9 * to_6
            mul_7: "bf16[s59, 32, 64]" = getitem_10 * to_7
            sub: "bf16[s59, 32, 64]" = mul_6 - mul_7;  mul_6 = mul_7 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_8: "bf16[s59, 32, 64]" = getitem_10 * to_6;  getitem_10 = to_6 = None
            mul_9: "bf16[s59, 32, 64]" = getitem_9 * to_7;  getitem_9 = to_7 = None
            add_5: "bf16[s59, 32, 64]" = mul_8 + mul_9;  mul_8 = mul_9 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat: "bf16[s59, 32, 128]" = torch.cat((sub, add_5), dim = -1);  sub = add_5 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:160 in forward_static, code: query = torch.cat((query_rot, query_pass), dim=-1).reshape(query_shape)
            cat_1: "bf16[s59, 32, 128]" = torch.cat((cat, getitem_8), dim = -1);  cat = getitem_8 = None
            reshape: "bf16[s59, 4096]" = cat_1.reshape(size);  cat_1 = size = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:164 in forward_static, code: key_shape = key.shape
            size_1 = getitem_3.size()
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:165 in forward_static, code: key = key.view(num_tokens, -1, head_size)
            view_2: "bf16[s59, 8, 128]" = getitem_3.view(s59, -1, 128);  getitem_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:166 in forward_static, code: key_rot = key[..., :rotary_dim]
            getitem_11: "bf16[s59, 8, 128]" = view_2[(Ellipsis, slice(None, 128, None))]
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:167 in forward_static, code: key_pass = key[..., rotary_dim:]
            getitem_12: "bf16[s59, 8, 0]" = view_2[(Ellipsis, slice(128, None, None))];  view_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:163 in forward_static, code: cos = cos.unsqueeze(-2).to(x.dtype)
            unsqueeze_2: "bf16[s59, 1, 64]" = getitem_5.unsqueeze(-2);  getitem_5 = None
            to_8: "bf16[s59, 1, 64]" = unsqueeze_2.to(torch.bfloat16);  unsqueeze_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:164 in forward_static, code: sin = sin.unsqueeze(-2).to(x.dtype)
            unsqueeze_3: "bf16[s59, 1, 64]" = getitem_6.unsqueeze(-2);  getitem_6 = None
            to_9: "bf16[s59, 1, 64]" = unsqueeze_3.to(torch.bfloat16);  unsqueeze_3 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:167 in forward_static, code: x1, x2 = torch.chunk(x, 2, dim=-1)
            chunk_2 = torch.chunk(getitem_11, 2, dim = -1);  getitem_11 = None
            getitem_13: "bf16[s59, 8, 64]" = chunk_2[0]
            getitem_14: "bf16[s59, 8, 64]" = chunk_2[1];  chunk_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:172 in forward_static, code: o1 = x1 * cos - x2 * sin
            mul_10: "bf16[s59, 8, 64]" = getitem_13 * to_8
            mul_11: "bf16[s59, 8, 64]" = getitem_14 * to_9
            sub_1: "bf16[s59, 8, 64]" = mul_10 - mul_11;  mul_10 = mul_11 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:173 in forward_static, code: o2 = x2 * cos + x1 * sin
            mul_12: "bf16[s59, 8, 64]" = getitem_14 * to_8;  getitem_14 = to_8 = None
            mul_13: "bf16[s59, 8, 64]" = getitem_13 * to_9;  getitem_13 = to_9 = None
            add_6: "bf16[s59, 8, 64]" = mul_12 + mul_13;  mul_12 = mul_13 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/common.py:176 in forward_static, code: output = torch.cat((o1, o2), dim=-1)
            cat_2: "bf16[s59, 8, 128]" = torch.cat((sub_1, add_6), dim = -1);  sub_1 = add_6 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/rotary_embedding/base.py:174 in forward_static, code: key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
            cat_3: "bf16[s59, 8, 128]" = torch.cat((cat_2, getitem_12), dim = -1);  cat_2 = getitem_12 = None
            reshape_1: "bf16[s59, 1024]" = cat_3.reshape(size_1);  cat_3 = size_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:414 in forward, code: output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
            size_2 = torch.Size([s59, 4096]);  s59 = None
            empty: "bf16[s59, 4096]" = torch.empty(size_2, dtype = torch.bfloat16, device = device(type='cuda', index=0));  size_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:419 in forward, code: query = query.view(-1, self.num_heads, self.head_size)
            view_3: "bf16[s59, 32, 128]" = reshape.view(-1, 32, 128);  reshape = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:420 in forward, code: output = output.view(-1, self.num_heads, self.head_size_v)
            view_4: "bf16[s59, 32, 128]" = empty.view(-1, 32, 128);  empty = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:422 in forward, code: key = key.view(-1, self.num_kv_heads, self.head_size)
            view_5: "bf16[s59, 8, 128]" = reshape_1.view(-1, 8, 128);  reshape_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:424 in forward, code: value = value.view(-1, self.num_kv_heads, self.head_size_v)
            view_6: "bf16[s59, 8, 128]" = getitem_4.view(-1, 8, 128);  getitem_4 = None
            return (view_5, view_6, view_3, view_4, to_4)
            
    class submod_115(torch.nn.Module):
        def forward(self, key_173: "bf16[s59, 8, 128]", s59: "Sym(s59)", value_57: "bf16[s59, 8, 128]", query_173: "bf16[s59, 32, 128]", output_231: "bf16[s59, 32, 128]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:453 in forward, code: kv_cache_dummy_dep = torch.ops.vllm.unified_kv_cache_update(
            unified_kv_cache_update: "bf16[0]" = torch.ops.vllm.unified_kv_cache_update(key_173, value_57, 'language_model.model.layers.25.self_attn.attn')
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:456 in forward, code: torch.ops.vllm.unified_attention_with_output(
            unified_attention_with_output = torch.ops.vllm.unified_attention_with_output(query_173, key_173, value_57, output_231, 'language_model.model.layers.25.self_attn.attn', kv_cache_dummy_dep = unified_kv_cache_update);  query_173 = key_173 = value_57 = output_231 = unified_kv_cache_update = unified_attention_with_output = None
            return ()
            
    class submod_116(torch.nn.Module):
        def forward(self, output_231: "bf16[s59, 32, 128]", s59: "Sym(s59)", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_: "bf16[3072, 4096]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_: "bf16[3072]", residual_49: "bf16[s59, 3072]", t_cond: "bf16[1, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_: "bf16[32, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_: "bf16[3072, 32]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_: "bf16[18432, 3072]", l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_: "bf16[3072, 9216]", l_self_modules_language_model_modules_model_modules_norm_parameters_weight_: "bf16[3072]"):
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/attention/attention.py:464 in forward, code: return output.view(-1, hidden_size)
            view: "bf16[s59, 4096]" = output_231.view(-1, 4096);  output_231 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear: "bf16[s59, 3072]" = torch._C._nn.linear(view, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_, None);  view = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_self_attn_modules_o_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_post_attention_layernorm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to: "f32[s59, 3072]" = linear.to(torch.float32);  linear = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add: "f32[s59, 3072]" = to + residual_49;  to = residual_49 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_1: "bf16[s59, 3072]" = add.to(torch.bfloat16)
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_1: "f32[s59, 3072]" = add.pow(2)
            mean: "f32[s59, 1]" = pow_1.mean(dim = -1, keepdim = True);  pow_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_1: "f32[s59, 1]" = mean + 1e-05;  mean = None
            rsqrt: "f32[s59, 1]" = torch.rsqrt(add_1);  add_1 = None
            mul: "f32[s59, 3072]" = add * rsqrt;  add = rsqrt = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_2: "bf16[s59, 3072]" = mul.to(torch.bfloat16);  mul = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_1: "bf16[s59, 3072]" = to_2 * _get_data_attr;  to_2 = _get_data_attr = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_1: "bf16[1, 32]" = torch._C._nn.linear(t_cond, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_, None);  t_cond = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_0_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            gelu: "bf16[1, 32]" = torch._C._nn.gelu(linear_1, approximate = 'none');  linear_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_2: "bf16[1, 3072]" = torch._C._nn.linear(gelu, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_, None);  gelu = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_ada_rms_norm_t_cond_modules_2_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/models/mistral.py:200 in forward, code: hidden_states = hidden_states * (1 + self.ada_rms_norm_t_cond(t_cond))
            add_2: "bf16[1, 3072]" = 1 + linear_2;  linear_2 = None
            mul_2: "bf16[s59, 3072]" = mul_1 * add_2;  mul_1 = add_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_3: "bf16[s59, 18432]" = torch._C._nn.linear(mul_2, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_, None);  mul_2 = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_gate_up_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py:141 in forward_native, code: return F.silu(x[..., :d]) * x[..., d:]
            getitem: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(None, 9216, None))]
            silu: "bf16[s59, 9216]" = torch.nn.functional.silu(getitem);  getitem = None
            getitem_1: "bf16[s59, 9216]" = linear_3[(Ellipsis, slice(9216, None, None))];  linear_3 = None
            mul_3: "bf16[s59, 9216]" = silu * getitem_1;  silu = getitem_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/parameter.py:126 in __torch_function__, code: return super().__torch_function__(func, types, args, kwargs)
            linear_4: "bf16[s59, 3072]" = torch._C._nn.linear(mul_3, l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_, None);  mul_3 = l_self_modules_language_model_modules_model_modules_layers_modules_25_modules_mlp_modules_down_proj_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:192 in forward_native, code: self.weight.data if self.has_weight else None,
            _get_data_attr_1: "bf16[3072]" = torch._C._autograd._get_data_attr(l_self_modules_language_model_modules_model_modules_norm_parameters_weight_);  l_self_modules_language_model_modules_model_modules_norm_parameters_weight_ = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:145 in forward_static, code: x = x.to(torch.float32)
            to_3: "f32[s59, 3072]" = linear_4.to(torch.float32);  linear_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:150 in forward_static, code: x = x + residual
            add_3: "f32[s59, 3072]" = to_3 + to_1;  to_3 = to_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:151 in forward_static, code: residual = x.to(orig_dtype)
            to_4: "bf16[s59, 3072]" = add_3.to(torch.bfloat16);  to_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:169 in forward_static, code: variance = x_var.pow(2).mean(dim=-1, keepdim=True)
            pow_2: "f32[s59, 3072]" = add_3.pow(2)
            mean_1: "f32[s59, 1]" = pow_2.mean(dim = -1, keepdim = True);  pow_2 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:171 in forward_static, code: x = x * torch.rsqrt(variance + variance_epsilon)
            add_4: "f32[s59, 1]" = mean_1 + 1e-05;  mean_1 = None
            rsqrt_1: "f32[s59, 1]" = torch.rsqrt(add_4);  add_4 = None
            mul_4: "f32[s59, 3072]" = add_3 * rsqrt_1;  add_3 = rsqrt_1 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:172 in forward_static, code: x = x.to(orig_dtype)
            to_5: "bf16[s59, 3072]" = mul_4.to(torch.bfloat16);  mul_4 = None
            
            # File: /home/ubuntu/mstral/venv/lib/python3.10/site-packages/vllm/model_executor/layers/layernorm.py:174 in forward_static, code: x = x * weight
            mul_5: "bf16[s59, 3072]" = to_5 * _get_data_attr_1;  to_5 = _get_data_attr_1 = None
            return mul_5