o ei(@sdZddlmZddlZddlmZddlmZddlm Z ddl m Z dd l m Z dd lmZd d lmZd d lmZmZmZmZmZmZmZddlmZeeZGdddeZGdddeZ GdddeZ!GdddeZ"GdddeZ#GdddeZ$gdZ%dS)zPyTorch BitNet model.)CallableN)Cache)FlashAttentionKwargs)CausalLMOutputWithPast)ALL_ATTENTION_FUNCTIONS)Unpack)logging)GemmaMLP)LlamaAttentionLlamaDecoderLayerLlamaForCausalLM LlamaModel LlamaRMSNormapply_rotary_pos_embeager_attention_forward) BitNetConfigc@ eZdZdS) BitNetRMSNormN__name__ __module__ __qualname__rrg/home/ubuntu/transcripts/venv/lib/python3.10/site-packages/transformers/models/bitnet/modular_bitnet.pyr)rcs*eZdZdeffdd ZddZZS) BitNetMLPconfigcs"t|t|j|jd|_dSN)eps)super__init__rintermediate_size rms_norm_eps ffn_sub_norm)selfr __class__rrr#.s zBitNetMLP.__init__c Cs*|||||||}|S)N) down_projr&act_fn gate_projup_proj)r'xr*rrrforward2s&zBitNetMLP.forward)rrrrr#r/ __classcell__rrr(rr-srcseZdZdedeffdd Z  ddejdeejejfdejdBd e dBd ej dBd e e d eejejdBffd dZ ZS)BitNetAttentionr layer_idxcs$t||t|j|jd|_dSr )r"r#r hidden_sizer% attn_sub_norm)r'rr2r(rrr#8szBitNetAttention.__init__N hidden_statesposition_embeddingsattention_maskpast_key_valuescache_positionkwargsreturncKs"|jdd}g|d|jR}|||dd} |||dd} |||dd} |\} } t| | | | \} } |durW| | |d}|| | |j |\} } t |j j t}||| | | |f|jskdn|j|jd|\}}|jg|dR}||}||}||fS)Nrr )sincosr9g)dropoutscaling)shapehead_dimq_projview transposek_projv_projrupdater2r get_interfacer_attn_implementationrtrainingattention_dropoutr@reshape contiguousr4o_proj)r'r5r6r7r8r9r: input_shape hidden_shape query_states key_states value_statesr>r= cache_kwargsattention_interface attn_output attn_weightsrrrr/<s:     zBitNetAttention.forward)NN)rrrrintr#torchTensortupler LongTensorrrr/r0rrr(rr17s& r1c@r)BitNetDecoderLayerNrrrrrr^irr^c@r) BitNetModelNrrrrrr_mrr_cs2eZdZddiZdZdZdeffdd ZZS)BitNetForCausalLMzlm_head.weightzmodel.embed_tokens.weightNr;c stjdi|S)a$ labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should either be in `[0, transformers., config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, transformers., config.vocab_size]`. Example: ```python >>> from transformers import AutoTokenizer, BitNetForCausalLM >>> model = BitNetForCausalLM.from_pretrained("microsoft/bitnet-b1.58-2B-4T") >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/bitnet-b1.58-2B-4T") >>> prompt = f'<|begin_of_text|>User: Hey, are you conscious? Can you talk to me?<|eot_id|>Assistant: ' >>> inputs = tokenizer(prompt, return_tensors="pt") >>> # Generate >>> generate_ids = model.generate(inputs.input_ids, max_length=100) >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] "User: Hey, are you conscious? Can you talk to me?Assistant: No, I'm not conscious. I'm an artificial intelligence designed to assist with information and tasks. How can I help you today?" ```Nr)r"r/)r' super_kwargsr(rrr/vszBitNetForCausalLM.forward) rrr_tied_weights_keys_tp_plan_pp_planrr/r0rrr(rr`qsr`)r`r_BitNetPreTrainedModel)&__doc__collections.abcrrZ cache_utilsrmodeling_flash_attention_utilsrmodeling_outputsrmodeling_utilsrprocessing_utilsrutilsr gemma.modeling_gemmar llama.modeling_llamar r rrrrrconfiguration_bitnetr get_loggerrloggerrrr1r^r_r`__all__rrrrs(        $  2 "