bCi`SrSSKJrJr SSKrSSKJr SSKJr SSK J r SSK J r SS K Jr SS KJr SS KJr S S KJr S SKJrJrJrJrJrJrJr SSKJr \R>"\ 5r!"SS\5r""SS\5r#"SS\5r$"SS\5r%"SS\5r&"SS\5r'/SQr(g)zPyTorch BitNet model.)CallableOptionalN)Cache)FlashAttentionKwargs)CausalLMOutputWithPast)ALL_ATTENTION_FUNCTIONS)Unpack)logging)deprecate_kwarg)GemmaMLP)LlamaAttentionLlamaDecoderLayerLlamaForCausalLM LlamaModel LlamaRMSNormapply_rotary_pos_embeager_attention_forward) BitNetConfigc\rSrSrSrg) BitNetRMSNorm+N__name__ __module__ __qualname____firstlineno____static_attributes__rc/home/james-whalen/.local/lib/python3.13/site-packages/transformers/models/bitnet/modular_bitnet.pyrr+r"rc6^\rSrSrS\4U4SjjrSrSrU=r$) BitNetMLP/configcj>[TU]U5 [URURS9UlgN)eps)super__init__rintermediate_size rms_norm_eps ffn_sub_norm)selfr( __class__s r#r-BitNetMLP.__init__0s+  )&*B*BH[H[\r"c URURURURU55UR U5-55nU$)N) down_projr0act_fn gate_projup_proj)r1xr5s r#forwardBitNetMLP.forward4sFNN4#4#4T[[PQAR5SVZVbVbcdVe5e#fg r")r0) rrrr rr-r:r! __classcell__r2s@r#r&r&/s]|]r"r&c6^\rSrSrS\S\4U4Sjjr\"SSSS9SS \RS \ \R\R4S \ \RS\ \ S \ \RS \\S\ \R\ \R44Sjj5rSrU=r$)BitNetAttention9r( layer_idxcj>[TU]X5 [URURS9Ulgr*)r,r-r hidden_sizer/ attn_sub_norm)r1r(rAr2s r#r-BitNetAttention.__init__:s+ +*6+=+=6CVCVWr"past_key_valuepast_key_valuesz4.58)new_nameversion hidden_statesposition_embeddingsattention_maskcache_positionkwargsreturnc VURSSn/UQSPURP7nURU5RU5R SS5n UR U5RU5R SS5n UR U5RU5R SS5n Uup[XX5upUb$XUS.nURXURU5up[nURRS:wa[URRnU"UU U U U4UR(dSO URUR S.UD6unnUR""/UQSP76R%5nUR'U5nUR)U5nUU4$)Nrr )sincosrMeagerg)dropoutscaling)shapehead_dimq_projview transposek_projv_projrupdaterArr(_attn_implementationr trainingattention_dropoutrVreshape contiguousrDo_proj)r1rJrKrLrGrMrN input_shape hidden_shape query_states key_states value_statesrSrR cache_kwargsattention_interface attn_output attn_weightss r#r:BitNetAttention.forward>s$))#2. 88b8$--8 {{=166|DNNqRST [[/44\BLLQPQR {{=166|DNNqRST &#7 RU#[  &#&nUL'6'='=jX\XfXfht'u $J(? ;; + +w 6"9$++:Z:Z"[ $7      % $}}C$2H2HLL %  % ! \"));;;;FFH ((5 kk+. L((r")rD)NN)rrrr rintr-r torchTensortuplerr LongTensorr rr:r!r<r=s@r#r?r?9sX|XX%0A6R ,059 +)||+)#5<<#=>+)!. +) "% +) !!1!12 +)-.+) u||Xell33 4+)S+)r"r?c\rSrSrSrg)BitNetDecoderLayermrNrrr"r#rurumr$r"ruc\rSrSrSrg) BitNetModelqrNrrr"r#rxrxqr$r"rxc>^\rSrSrS/rSrSrS\4U4SjjrSr U=r $)BitNetForCausalLMuzlm_head.weightNrOc $>[TU]"S0UD6$)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,r:)r1 super_kwargsr2s r#r:BitNetForCausalLM.forwardzs4w...r"r) rrrr _tied_weights_keys_tp_plan_pp_planrr:r!r<r=s@r#r{r{us**+HH/ //r"r{)r{rxBitNetPreTrainedModel))__doc__typingrrrp cache_utilsrmodeling_flash_attention_utilsrmodeling_outputsrmodeling_utilsr processing_utilsr utilsr utils.deprecationr gemma.modeling_gemmarllama.modeling_llamarrrrrrrconfiguration_bitnetr get_loggerrloggerrr&r?rurxr{__all__rr"r#rs% B65&0+/   H % L 1)n1)h *  * /(/D r"