+ j6Rt^RIHt^RIt^RIHt^RIHt^RIHt^RI H t ^RI H t H t ^RIHt^R IHt^R IHt^R IHtHtHtHt^R IHtHt^R IHtHtHtH t H!t!H"t"^RI#H$t$]PJ!]&4t'!RR]PP4t)!RR]!4t*Rt+RRlt,!RR]4t-]!RR]44t.!RR] 4t/!RR] 4t0!RR]4t1.R Ot2R#)!zPyTorch Cohere model.)CallableN)nn)Cache)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)dynamic_rope_update)ALL_ATTENTION_FUNCTIONS)Unpack)TransformersKwargsauto_docstringcan_return_tuplelogging)maybe_autocastno_inherit_decorator)LlamaAttentionLlamaForCausalLMLlamaMLP LlamaModelLlamaRotaryEmbeddingeager_attention_forward CohereConfigc<aa]tRt^4toRV3RlltRtRtVtV;t#)CohereLayerNormc <\SV`4\P!\P !V44VnW nR#)zcThe hidden size can be a tuple or an int. The tuple is used for QKNorm to normalize across head_dimN)super__init__r Parametertorchonesweightvariance_epsilon)self hidden_sizeepsbias __class__s&&&&j/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/cohere/modular_cohere.pyrCohereLayerNorm.__init__5s- ll5::k#:; #cVPpVP\P4pVP RRR7pW, P ^4P RRR7pW, \P !W@P,4,pVPP\P4V,pVPV4#)T)keepdim) dtypetor float32meanpowrsqrtr#r")r$ hidden_states input_dtyper3variances&& r)forwardCohereLayerNorm.forward;s#)) %((7 !!"d!3!(--a055b$5G&-XH]H]=]1^^  u}}5 E  ,,r+)r#r")Ngh㈵>F __name__ __module__ __qualname____firstlineno__rr9__static_attributes____classdictcell__ __classcell__r( __classdict__s@@r)rr4s$ --r+rcXa]tRt^Eto]P !4]R44tRtVt R#)CohereRotaryEmbeddingcbVPR,P4PVP^,R^4pVR ,P4p\ VP P \4'd2VP P R8wdVP P MRp\VRR7;_uu_4VP4VP4,P^^4p\P!V^RR7pVP4VP,pVP4VP,p RRR4XPVP R7X PVP R73# +'giLH;i) NmpscpuF) device_typeenableddimr0)NNNNNr/)rONrO)inv_freqfloatexpandshape isinstancedevicetypestrr transposer repeat_interleavecosattention_scalingsinr1r0) r$x position_idsinv_freq_expandedposition_ids_expandedrJfreqsembrZr\s &&& r)r9CohereRotaryEmbedding.forwardFs5!MM-8>>@GG HZHZ[\H]_acde ,Z 8 > > @'1!((--'E'E!((--[`J`ahhmmfk  U C C&,,.1F1L1L1NNYYZ[]^_E))%;C'')d444C'')d444C D vvAGGv$cff177f&;;; D Cs BF F. N) r<r=r>r?r no_gradr r9r@rA)rDs@r)rFrFEs% ]]_ < C/0Gw;q>C/0G ::E: "GJJUJ$; ;;r+c2aa]tRt^ztoV3RltRtVtV;t#) CohereMLPcR<\SV`V4\P!VPVP RR7Vn\P!VPVP RR7Vn\P!VP VPRR7VnR#)F)r'N) rrrLinearr%intermediate_size gate_projup_proj down_projr$configr(s&&r)rCohereMLP.__init__{ss  4#3#3T5K5KRWXyy!1!143I3IPUV 4#9#94;K;KRWXr+)r~r|r}r<r=r>r?rr@rArBrCs@@r)rxrxzsYYr+rxc\aa]tRt^toRtRV3RlV3RllltRV3RlRlltRtVtV;t #)CohereAttentionz=Multi-headed attention from 'Attention Is All You Need' paperc4<V^8dQhRS[RS[R,/#)rgr layer_idxNrint)formatrDs"r) __annotate__CohereAttention.__annotate__s  | d r+c8<\SV`W4VPVnVP'dg\VPVP 3VP R7Vn\VPVP 3VP R7Vn R#R#r%r&N) rr use_qk_normrnum_attention_headshead_dimlayer_norm_epsq_normnum_key_value_headsk_normr$rrr(s&&&r)rCohereAttention.__init__s} +!--    )#77GVMbMbDK*#77GVMbMbDK r+c<V^8dQhRS[PRS[S[PS[P3,RS[PR,RS[R,RS[S[,RS[S[PS[PR,3,/#)rgr6position_embeddingsattention_maskNpast_key_valueskwargsreturn)r Tensortuplerr r)rrDs"r)rrs.).)||.)#5<<#=>.) t+ .)  .) -. .) u||U\\D00 1.)r+c VPRRp.VORNVPN5pVPV4PV4pVP V4PV4p VP V4PV4p VP 'd#VPV4pVPV 4p VP^^4pV P^^4p V P^^4p Vwr\WW4wrVeVPWVP4wr\P!VPP \"4p T !VVV V V3RVP$'gRM VP&RVP(/VBwrVP*!.VORN5!P-4pVP/V4pW3#)Ndropoutgscalingr/)rSrq_projviewk_projv_projrrrrXrvupdaterr get_interfacer_attn_implementationrtrainingattention_dropoutrreshape contiguouso_proj)r$r6rrrr input_shape hidden_shape query_states key_states value_statesrZr\attention_interface attn_output attn_weightss&&&&&, r)r9CohereAttention.forwards$))#2. 88b8$--8 {{=166|D [[/44\B {{=166|D    ;;|4LZ0J#--a3 ))!Q/ #--a3 &#7 RU#[  &'6'='=jX\XfXf'g $J(?(M(M KK , ,.E) %8      % $}}}C$2H2H % LL %  % ! "));;;;FFH kk+. ((r+)rrrN) r<r=r>r?__doc__rr9r@rArBrCs@@r)rrs!G  .).).)r+rcTaa]tRt^toV3RlV3RlltRV3RlRlltRtVtV;t#)CohereDecoderLayerc&<V^8dQhRS[RS[/#)rgrrr)rrDs"r)rCohereDecoderLayer.__annotate__sll|llr+c<\SV`4VPVn\WR7Vn\ V4Vn\VPVPR7Vn R#))rrrN) rrr%r self_attnrxmlprrinput_layernormrs&&&r)rCohereDecoderLayer.__init__sP !--(LV$.FE*-.* *r+c TpVPV4pVP!RRVRVRVRVRVRV/VBwrVPV4p W,V ,pV#)a< Args: hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)` attention_mask (`torch.FloatTensor`, *optional*): attention mask of size `(batch_size, sequence_length)` if flash attention is used or `(batch_size, 1, query_sequence_length, key_sequence_length)` if default attention is used. past_key_values (`Cache`, *optional*): cached past key and value projection states output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. use_cache (`bool`, *optional*): If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see `past_key_values`). position_embeddings (`tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*): Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`, with `head_dim` being the embedding dimension of each attention head. r6rr^rrrrd)rrr) r$r6rr^rrrrresidualhidden_states_attention_hidden_states_mlps &&&&&&&, r)r9CohereDecoderLayer.forwards6!,,]; %)^^& '& )& && , &  & !4 & & "!HH]3 :=NN r+)r%rrr)NNNFNr;rCs@@r)rrs ll***r+rc>aa]tRt^toV3RlV3RlltRtVtV;t#) CohereModelc <V^8dQhRS[/#)rgrr)rrDs"r)rCohereModel.__annotate__saa|ar+c  <\SV`V4\P!\ VP 4Uu.uFp\ W4NK up4Vn\VPVPR7Vn R#uupir) rrr ModuleListrangenum_hidden_layersrlayersrr%rnormrs&& r)rCohereModel.__init__sg  mmDI&JbJbDc dDcy  2Dc d  $1C1C&J_J_`  esB)rrrrCs@@r)rrsaaar+rc\aa]tRtRtoV3Rlt]]RV3RlRll44tRtVt V;t #)CohereForCausalLMic<\SV`V4\V4VnVPVnVP VnR#r)rrrmodel logit_scaletie_word_embeddingsrs&&r)rCohereForCausalLM.__init__s8   ( !--#)#=#= r+cP<V^8dQhRS[PR,RS[PR,RS[PR,RS[R,RS[PR,RS[PR,RS[R,R S[S[P,R S[S[,R S[ / #) rg input_idsNrr^r inputs_embedslabelsrlogits_to_keeprr) r rrr FloatTensorrrr r r)rrDs"r)rCohereForCausalLM.__annotate__ s6 6 ##d*6  t+6 &&- 6  6 ((4/ 6   4'6 $;6 ell*6 +,6  6 r+c  VP!R RVRVRVRVRVRV/V Bp V Pp \V\4'd\ V)R4MTp VP V RV R3,4p WP ,p RpVe-VP!R R V R VR VPP/V Bp\VV V PV PV PR 7#)au Example: ```python >> from transformers import AutoTokenizer, CohereForCausalLM >> model = CohereForCausalLM.from_pretrained("CohereForAI/c4ai-command-r-v01") >> tokenizer = AutoTokenizer.from_pretrained("CohereForAI/c4ai-command-r-v01") >> prompt = "Hey, are you conscious? Can you talk to me?" >> inputs = tokenizer(prompt, return_tensors="pt") >> # Generate >> generate_ids = model.generate(inputs.input_ids, max_length=30) >> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you." ```rrr^rrrNrOlogitsr vocab_size)lossrrr6 attentionsrd)rlast_hidden_staterTrslicelm_headr loss_functionrrrrr6r)r$rrr^rrrrrroutputsr6 slice_indicesrrs&&&&&&&&&, r)r9CohereForCausalLM.forwards >,0::, , ), &, , , ( ,  , ,  11 8B>SV8W8W~ot4]k mA}a,?@A***  %%pVpFpt{{OeOepiopD%#33!//))   r+)rrr)NNNNNNN) r<r=r>r?rrr r9r@rArBrCs@@r)rrs0> 6 6 6 6 r+r)r-)rrCoherePreTrainedModel)3rcollections.abcrr r cache_utilsrmodeling_flash_attention_utilsrmodeling_layersrmodeling_outputsrrmodeling_rope_utilsr modeling_utilsr processing_utilsr utilsr r rr utils.genericrrllama.modeling_llamarrrrrrconfiguration_coherer get_loggerr<loggerModulerrFrnrvrxrrrr__all__rdr+r)rs,$  B9O65&RRA/   H %-bii-"<0<"<8YY=)n=)=)@232ja*a? (? D r+