o پir @sddlmZmZddlmZmZddlZddlmZddl m Z e e Z ededZ Gdd d eee Zd ejeejBd eedBd ejjdBd edejf ddZdS))ABCabstractmethod)GenericTypeVarN)PretrainedConfig) init_logger_C)boundcseZdZdeddffdd Zedededefdd Zedefd d Zedefd d Z edefddZ edefddZ Z S)VisionEncoderInfo vision_configreturnNcst||_dSN)super__init__r )selfr  __class__h/home/ubuntu/.local/lib/python3.10/site-packages/sglang/multimodal_gen/runtime/models/encoders/vision.pyrs  zVisionEncoderInfo.__init__ image_width image_heightcCtr NotImplementedError)rrrrrrget_num_image_tokenssz&VisionEncoderInfo.get_num_image_tokenscCrr rrrrrget_max_image_tokens#z&VisionEncoderInfo.get_max_image_tokenscCrr rrrrrget_image_size'rz VisionEncoderInfo.get_image_sizecCrr rrrrrget_patch_size+rz VisionEncoderInfo.get_patch_sizecCrr rrrrrget_patch_grid_length/rz'VisionEncoderInfo.get_patch_grid_length) __name__ __module__ __qualname__rrrintrrrrr __classcell__rrrrr s$ r encoder_outputsfeature_sample_layerspost_layer_normmax_possible_layersr cs|dur|dur |SStd}||fdd|D}|dt|ddfv}|dur:|r:||d<tj|ddS)a^Given the outputs a visual encoder module that may correspond to the output of the last layer, or a list of hidden states to be stacked, handle post normalization and resolve it into a single output tensor. Args: encoder_outputs: Output of encoder's last layer or all hidden states. feature_sample_layers: Optional layer indices to grab from the encoder outputs; if provided, encoder outputs must be a list. post_layer_norm: Post norm to apply to the output of the encoder. max_possible_layers: Total layers in the fully loaded visual encoder. Ncs(g|]}|dkr |n|qS)rr).0 layer_idxr&offsetrr Ss  z2resolve_visual_encoder_outputs..)dim)lentorchcat)r&r'r(r)num_loaded_layershs_pooluses_last_layerrr-rresolve_visual_encoder_outputs4s    r8)abcrrtypingrrr3 transformersr1sglang.multimodal_gen.runtime.utils.logging_utilsrr!loggerrr Tensorlistr$nn LayerNormr8rrrrs&   !