i*PdZddlmZddlZddlmZddlmZddlm Z ddl m Z dd l m Z dd lmZd d lmZd d lmZmZmZmZmZmZmZddlmZej8eZGddeZGddeZ GddeZ!GddeZ"GddeZ#GddeZ$gdZ%y)zPyTorch BitNet model.)CallableN)Cache)FlashAttentionKwargs)CausalLMOutputWithPast)ALL_ATTENTION_FUNCTIONS)Unpack)logging)GemmaMLP)LlamaAttentionLlamaDecoderLayerLlamaForCausalLM LlamaModel LlamaRMSNormapply_rotary_pos_embeager_attention_forward) BitNetConfigc eZdZy) BitNetRMSNormN__name__ __module__ __qualname__z/var/www/vps2.regionflexible.com/Desarrollo/venv/lib/python3.12/site-packages/transformers/models/bitnet/modular_bitnet.pyrr)rrc*eZdZdeffd ZdZxZS) BitNetMLPconfigcpt||t|j|j|_yN)eps)super__init__rintermediate_size rms_norm_eps ffn_sub_norm)selfr" __class__s rr'zBitNetMLP.__init__.s+  )&*B*BH[H[\rc |j|j|j|j||j |z}|SN) down_projr*act_fn gate_projup_proj)r+xr/s rforwardzBitNetMLP.forward2sFNN4#4#4T[[PQAR5SVZVbVbcdVe5e#fg r)rrrrr'r4 __classcell__r,s@rr!r!-s]|]rr!ceZdZdedeffd Z d dejdeejejfdejdzde dzd e e d eejejdzff d Z xZ S) BitNetAttentionr" layer_idxcrt|||t|j|j|_yr$)r&r'r hidden_sizer) attn_sub_norm)r+r"r9r,s rr'zBitNetAttention.__init__8s- +*6+=+=6CVCVWrN hidden_statesposition_embeddingsattention_maskpast_key_valueskwargsreturnc ,|jdd}g|d|j}|j|j|j dd}|j |j|j dd} |j |j|j dd} |\} } t|| | | \}} | |j| | |j\} } tj|jjt} | ||| | |f|jsdn |j |j"d|\}}|j$g|dj'}|j)|}|j+|}||fS)Nrr g)dropoutscaling)shapehead_dimq_projview transposek_projv_projrupdater9r get_interfacer"_attn_implementationrtrainingattention_dropoutrFreshape contiguousr<o_proj)r+r=r>r?r@rA input_shape hidden_shape query_states key_states value_statescossinattention_interface attn_output attn_weightss rr4zBitNetAttention.forward<s$))#2. 88b8$--8 {{=166|DNNqRST [[/44\BLLQPQR {{=166|DNNqRST &S#7 jRUWZ#[ j  &'6'='=j,X\XfXf'g $J (?(M(M KK , ,.E) %8      % $}}C$2H2HLL %  % ! \*k));;;;FFH ((5 kk+. L((rr.)rrrrintr'torchTensortuplerr rr4r5r6s@rr8r87sX|XX)- ')||')#5<<#=>') t+ ')  ') -. ') u||U\\D00 1')rr8c eZdZy)BitNetDecoderLayerNrrrrrerefrrrec eZdZy) BitNetModelNrrrrrgrgjrrrgc4eZdZddiZdZdZdeffd ZxZS)BitNetForCausalLMzlm_head.weightzmodel.embed_tokens.weightNrBc "t|di|S)a$ 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?" ```r)r&r4)r+ super_kwargsr,s rr4zBitNetForCausalLM.forwardss4w...r) rrr_tied_weights_keys_tp_plan_pp_planrr4r5r6s@rririns-*,GHHH/ //rri)rirgBitNetPreTrainedModel)&__doc__collections.abcrra cache_utilsrmodeling_flash_attention_utilsrmodeling_outputsrmodeling_utilsrprocessing_utilsr utilsr gemma.modeling_gemmar llama.modeling_llamar rrrrrrconfiguration_bitnetr get_loggerrloggerrr!r8rergri__all__rrrr~s$ B65&+/   H % L ,)n,)^ *  * /(/D r