ZjbbSrSSKrSSKJr SSKrSSKrSSKJr SSKJ r SSK J r SSK Jr SS KJrJrJr SS KJr SS KJrJr SS KJr SS KJr SSKJrJrJrJ r J!r! SSK"J#r#J$r$ SSK%J&r& SSK'J(r(J)r)J*r*J+r+J,r, SSK-J.r. SSK/J0r0J1r1 SSK2J3r3 \,Rh"\55r6S\RnS\8S\84Sjr9"SS\Rt5r;S"S%S&\Rx5r?"S'S(\5r@"S)S*\5rA\)"S+S,\$55rB"S-S.\B5rC"S/S0\B5rD\)"S1S2\B55rE\)"S3S49"S5S6\B\55rF"S7S8\B5rG"S9S:\B\5rH/S;QrIg)=z=PyTorch MarianMTModel model, ported from the Marian C++ repo.N)Callable)nn)CrossEntropyLoss)initialization)ACT2FN)Cache DynamicCacheEncoderDecoderCache)GenerationMixin)create_bidirectional_maskcreate_causal_mask)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutput)BaseModelOutputWithPastAndCrossAttentions!CausalLMOutputWithCrossAttentionsSeq2SeqLMOutputSeq2SeqModelOutput)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstringcan_return_tupleis_torchdynamo_compilinglogging)merge_with_config_defaults)OutputRecordercapture_outputs) MarianConfig input_ids pad_token_iddecoder_start_token_idcURUR5nUSS2SS24R5USS2SS24'X#SS2S4'Uc [S5eUR US:HU5 U$)z) Shift input ids one token to the right. Nr!rz1self.model.config.pad_token_id has to be defined.i) new_zerosshapeclone ValueError masked_fill_)r#r$r%shifted_input_idss {/root/GenerationalWealth/GenerationalWealth/venv/lib/python3.13/site-packages/transformers/models/marian/modeling_marian.pyshift_tokens_rightr/8sz"++IOO<(CRC0668ae4adLMM""#4#[TU]XSS9 g)NT)_freeze)super__init__)selfr4r5r6 __class__s r.r;,MarianSinusoidalPositionalEmbedding.__init__Ks tDr0cURRup[R"[ U5VVs/sH?n[ U5Vs/sH%oC[R "SSUS--U- 5- PM' snPMA snn5n[ R"XURRSS9nUS-S:XaUS-OUS-S-n[ R"[R"USS2SSS2455USS2SU24'[ R"[R"USS2SSS2455USS2US24'U$s snfs snnf)z Identical to the XLM create_sinusoidal_embeddings except features are not interleaved. The cos features are in the 2nd half of the vector. [dim // 2:] i'F)dtype requires_gradrr!N) weightr)nparrayrangepowertorchemptyrA FloatTensorsincos)r<n_posdimposj position_encoutsentinels r. create_weight1MarianSinusoidalPositionalEmbedding.create_weightNs [[&& xxX]^cXd eXdQTs LABHHUAaL3$677 LXd e kk%DKK,=,=US"Qw!|3!8#(a"..rvvl1add76K/LMAqzM!--bff\!QTT'5J.KLAxyL M esE ,E6E E input_ids_shapepast_key_values_length position_idsc>UcBUSSupE[R"X"U-[RURRS9n[ TU]U5$)z3`input_ids_shape` is expected to be [bsz x seqlen].Nr@)rAdevice)rHarangelongrCrZr:forward)r<rVrWrXbszseq_lenr=s r.r]+MarianSinusoidalPositionalEmbedding.forward]sX  *2A.LC <<&(HPUPZPZcgcncncucuLw|,,r0N)rN)__name__ __module__ __qualname____firstlineno____doc__intr;rTrHno_gradSizeTensorr]__static_attributes__ __classcell__r=s@r.r2r2HsNEcE#ECRVJEbfEE  ]]_pt -$zz -CF -Z_ZfZfimZm -  - -r0r2modulequerykeyvalueattention_maskscalingdropoutkwargsc UcURS5S-n[R"XRSS55U-nUbX-n[R R USS9n[R RXURS9n[R"X5n U RSS5R5n X4$)Nr'r@rrNptrainingr!) sizerHmatmul transposer functionalsoftmaxrur| contiguous) rorprqrrrsrtrurv attn_weights attn_outputs r.eager_attention_forwardrks**R.D(<<}}Q':;gEL!#4 ==((2(>L==((6??([L,,|3K''1-88:K  $$r0c$^\rSrSrSrSS\S\S\S\S\S \S \S-S \S-4U4S jjjr SS \ RS\ RS-S\ S-S\ RS-S\ \S\\ R\ RS-44 SjjrSrU=r$)MarianAttentionz=Multi-headed attention from 'Attention Is All You Need' paperN embed_dim num_headsru is_decoderbias is_causalconfig layer_idxc t>[T U]5 XlX lX0lX-UlXplUR U-UR:wa[SURSUS35eUR S-UlX@l X`l Xl Uc>UR(a-[RSURRS35 [ R""XUS9Ul[ R""XUS9Ul[ R""XUS9Ul[ R""XUS9Ulg)Nz;embed_dim must be divisible by num_heads (got `embed_dim`: z and `num_heads`: z).rxzInstantiating a decoder z without passing `layer_idx` is not recommended and will lead to errors during the forward call, if caching is used. Please make sure to provide a `layer_idx` when creating this class.r)r:r;rrruhead_dimrr+rtrrrlogger warning_oncer=rcrLineark_projv_projq_projout_proj) r<rrrurrrrrr=s r.r;MarianAttention.__init__s  "" !.  MMI %$.. 8MdnnM]$YKr3 }}d* $""     *4>>+B+B*CD,,  ii 4@ ii 4@ ii 4@  )TB r0 hidden_stateskey_value_statespast_key_valuesrsrvr7c ,USLnURSSn/UQSPURP7nURU5RU5R SS5n Sn Ub][ U[ 5(aFURRUR5n U(a URn OURn OUn U(aUOUn U(aQUbNU (aGW RURRn U RURRnOURU 5n UR!U 5n/U RSSQSPURP7nU RU5R SS5n URU5R SS5nUbSW R#XUR5upU(a.[ U[ 5(aSURUR'[$R&"UR(R*[,5nU"UU U UU4UR.(dSO UR0UR2S.UD6unnUR4"/UQSP76R75nUR9U5nUU4$) z#Input shape: Batch x Time x ChannelNr'r!r@FT)rurt)r)rrviewr isinstancer is_updatedgetrcross_attention_cacheself_attention_cachelayerskeysvaluesrrupdater get_interfacer_attn_implementationrr|rurtreshaperr)r<rrrrsrvis_cross_attention input_shape hidden_shape query_statesrcurr_past_key_valuescurrent_states key_states value_stateskv_shapeattention_interfacerrs r.r]MarianAttention.forwardsc.T9$))#2. 88b8$--8 {{=166|DNNqRST   &/+>??,77;;DNNK %+:+P+P(+:+O+O('6$-?)] /"=*-44T^^DIIJ/66t~~FMML^4J;;~6LF--cr2FBF FH#2<(?(M(M KK , ,.E) %8      % $}}C$,,LL %  % ! \"));;;;FFH mmK0 L((r0) rrurrrrrrrrrrtr)rFTFNNNNN)rcrdrerfrgrhfloatboolr"r;rHrkr rrtupler]rlrmrns@r.rrsG  &* $%C%C%C %C  %C  %C%Ct#%C:%C%CT15(,.2 H)||H) ,,-H) H)  t+ H)-.H) u||U\\D00 1H)H)r0rc^\rSrSrS S\S\S-4U4SjjjrS\RS\RS\ \ S \R4S jr S r U=r$) MarianEncoderLayerNrrcl>[TU]5 URUl[ URUR UR UUS9Ul[R"UR5Ul URUl [URUlURUl[R "URUR"5Ul[R "UR"UR5Ul[R"UR5Ulg)N)rrrurr)r:r;d_modelrrencoder_attention_headsattention_dropout self_attnr LayerNormself_attn_layer_normruractivation_function activation_fnactivation_dropoutrencoder_ffn_dimfc1fc2final_layer_normr<rrr=s r.r;MarianEncoderLayer.__init__s (nn44,,  %'LL$@!~~ #F$>$>?"(";";99T^^V-C-CD99V33T^^D " T^^ <r0rrsrvr7c 2UnUR"U4SU0UD6up[RRXRURS9nXA-nUR U5nUnUR URU55n[RRXRURS9nURU5n[RRXRURS9nXA-nURU5nUR[R:Xal[R"U5R5(dC[R "UR5R"S- n[R$"X*US9nU$)Nrsrzi)minmax)rrrrur|rrrrrrrArHfloat16isfiniteallfinforclamp)r<rrsrvresidual_ clamp_values r.r]MarianEncoderLayer.forwards[ !>>  )    --m||VZVcVc-d  0 11-@  **488M+BC  --m?V?Vaeanan-o /  --m||VZVcVc-d  0 --m<   %-- /}8U8Y8Y8[8[++m&9&9:>>EK!KK <[YMr0) rrrurrrrrrrb)rcrdrerfr"rhr;rHrJrrrkr]rlrmrns@r.rrse=|=d ==&(())+,   r0rc^\rSrSrSS\S\S-4U4SjjjrSS\RS\RS-S\RS-S \RS-S \ S-S \ S-S \ \ S \R4Sjjr SrU=r$)MarianDecoderLayeri2Nrrc >[TU]5 URUl[ URUR UR SSUUS9UlURUl[URUl URUl [R"UR5Ul[ URUR UR SUUS9Ul[R"UR5Ul[R$"URUR&5Ul[R$"UR&UR5Ul[R"UR5Ulg)NT)rrrurrrr)rurrr)r:r;rrrdecoder_attention_headsrrrurrrrrrr encoder_attnencoder_attn_layer_normrdecoder_ffn_dimrrrrs r.r;MarianDecoderLayer.__init__3s  (nn44,, ~~ #F$>$>?"(";";$&LL$@!+ NN  * *,,  (*||DNN'C$99T^^V-C-CD99V33T^^D " T^^ <r0rrsencoder_hidden_statesencoder_attention_maskr use_cachervr7c UnUR"U4UUS.UD6up[RRXRURS9nX-nUR U5nUbbUnUR "U4UUUS.UD6up[RRXRURS9nX-nURU5nUnURURU55n[RRXRURS9nURU5n[RRXRURS9nX-nURU5nU$)N)rrsrz)rrsr) rrrrur|rrrrrrrr) r<rrsrrrrrvrrs r.r]MarianDecoderLayer.forwardRs! >>  +)     --m||VZVcVc-d  0 11-@  ! ,$H#00 !65 /      MMM11-<[TU]U5 [U[5(a0[R "UR UR55 g[U[5(a![R"UR5 ggrb) r: _init_weightsrr2initcopy_rCrT MarianMTModelzeros_final_logits_bias)r<ror=s r.r#MarianPreTrainedModel._init_weightss] f% fA B B JJv}}f&:&:&< =  . . KK00 1/r0cURRn[R"/SQSSSSU//URS9nUR U5UUS.nU$)N)r r@r r@rZ)rsr#decoder_input_ids)rr$rHtensorrZne)r< pad_tokenr# dummy_inputss r.r"MarianPreTrainedModel.dummy_inputssZKK,, LL"2Q2q)4L!MVZVaVab 'll95"!* r0ra)rcrdrerfr"__annotations__base_model_prefixsupports_gradient_checkpointing_supports_flash_attn_supports_sdpa_supports_flex_attn_can_compile_fullgraphrHrirpropertyrrlrmrns@r.rrsX &*#N! ]]_22r0rc^\rSrSrSr\\S.rS\4U4Sjjr \ \ \ SS\ RS-S\ RS-S \ RS-S \\S \4 S jj555rS rU=r$) MarianEncoderiz Transformer encoder consisting of *config.encoder_layers* self attention layers. Each layer is a [`MarianEncoderLayer`]. Args: config: MarianConfig embed_tokens (nn.Embedding): output embedding )r attentionsrc>[TU]U5 URUlURUlUR nUR UlURUl UR(a[R"U5OSUl [R"UR X R5Ul[%URX R5Ul[R("[+UR,5Vs/sHn[/U5PM sn5UlSUlUR55 gs snf)N?F)r:r;ruencoder_layerdrop layerdroprr$r6max_position_embeddingsmax_source_positionsscale_embeddingmathsqrt embed_scaler Embedding vocab_size embed_tokensr2embed_positions ModuleListrFencoder_layersrrgradient_checkpointing post_init)r<rrrr=s r.r;MarianEncoder.__init__s  ~~ 11NN !..$*$B$B!393I3I499Y/sLL):):IGWGWXB  * *I7G7G mmvOdOdIe$fIeA%7%?Ie$fg &+#  %gs ENr#rs inputs_embedsrvr7c 0USLUSL- (a [S5eUcURU5UR-nURURSS5nX5-n[ R RX`RURS9n[URUUS9n[UR5HRupxSn UR(a'[R"/5n XR:aSn U (aMHU"UU40UD6nMT [!US9$)Nz:You must specify exactly one of input_ids or inputs_embedsr'rz)rrrsFT)last_hidden_state)r+rrrr)rrrur|r r enumeraterrHrandrr) r<r#rsrrv embed_posridx encoder_layerto_dropdropout_probabilitys r.r]MarianEncoder.forwards  -t"; <YZ Z   --i84;K;KKM(()<)[TU]U5 URUlURUlUR UlURUlUR(a [R"UR5OSUl [R"UR URUR 5Ul[%URURUR 5Ul[R("[+UR,5Vs/sH n[/XS9PM sn5UlSUlUR55 gs snf)Nr )rF)r:r;rudecoder_layerdroprr$r6rmax_target_positionsrrrrrrrdecoder_vocab_sizerr2rrrFdecoder_layersrrrr)r<rir=s r.r;MarianDecoder.__init__s  ~~ 11!..$*$B$B!8>8N8N499V^^4TWLL)B)BFNNTXTdTdeB  * *FNND??  0 0   );;')+  ";;;'1"7 " ++  $&<, &4  --mt||VZVcVc-d "+DKK"8 C}}&+jjn#&7)%(> /# M#9 9++  r0) rurrrrrrr2r6)NNNNNNN)rcrdrerfrgrrrr)r"r;rr rrHr*rkrJr rrrrr]rlrmrns@r.r,r,s,$_A+V*?!P^_ |& .2.2:>:>(,26!%R ##d*R  t+R  %0047 R !& 0 04 7 R  R ((4/R $;R +,R  3R  R r0r,c^\rSrSrSS/rS\4U4SjjrSrSrSr S r S \ S \ R4S jr\\SS\R$S -S\R&S -S\R$S -S\R&S -S\\R&\-S -S\S -S\R.S -S\R.S -S\S -S\\S \4Sjj55rSrU=r$) MarianModelip$model.encoder.embed_positions.weight$model.decoder.embed_positions.weightrcX>[TU]U5 URURp2URR (a1[ R"X1RU5Ul SSS.Ul OSUl [U5Ul [U5UlUR5 g)Nz shared.weight)zdecoder.embed_tokens.weightzencoder.embed_tokens.weight)r:r;r$rr share_encoder_decoder_embeddingsrrrshared_tied_weights_keysr encoderr,decoderr)r<rr6rr=s r.r;MarianModel.__init__ws  "("5"5v7H7HZ ;; 7 7,,z>>;ODK/>/>'D # '+D #$V, $V,  r0c>UR5R5$rb) get_encoderget_input_embeddingsr<s r.rO MarianModel.get_input_embeddingss!6688r0cURR(a=XlURURlURUR lgXRlgrb)rrGrHrJrrKr<rrs r.set_input_embeddings MarianModel.set_input_embeddingss> ;; 7 7K(, DLL %(, DLL %(-LL %r0cURR(a [S5eUR5R 5$)Nz`get_decoder_input_embeddings` should not be called if `config.share_encoder_decoder_embeddings` is `True`. Please use `get_input_embeddings` instead.)rrGr+ get_decoderrOrPs r.get_decoder_input_embeddings(MarianModel.get_decoder_input_embeddingss< ;; 7 7H !6688r0cpURR(a [S5eXRlg)Na `config.share_encoder_decoder_embeddings` is set to `True` meaning the decoder input embeddings are shared with the encoder. In order to set the decoder input embeddings, you should simply set the encoder input embeddings by calling `set_input_embeddings` with the appropriate embeddings.)rrGr+rKrrSs r.set_decoder_input_embeddings(MarianModel.set_decoder_input_embeddingss. ;; 7 7r  %* !r0new_num_tokensr7c URR(a [S5eUR5nUR X!5nUR U5 UR5nUcU$XRlUR5 U$Nz`resize_decoder_token_embeddings` should not be called if `config.share_encoder_decoder_embeddings` is `True`. Please use `resize_token_embeddings` instead.)rrGr+rX_get_resized_embeddingsr[r3 tie_weights)r<r]old_embeddingsnew_embeddings model_embedss r.resize_decoder_token_embeddings+MarianModel.resize_decoder_token_embeddingss ;; 7 7K  ::<55nU )).988:  ! *8 & r0Nr#rsrdecoder_attention_maskencoder_outputsrrdecoder_inputs_embedsrrvc UcUR"S UUUS.U D6nOK[U[5(d6[US[U5S:aUSOS[U5S:aUSOSS9nUR"S UUUSUUUU S.U D6n [ U R U RU RU RU RUR URURS9$) a decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*): Indices of decoder input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are decoder input IDs?](../glossary#decoder-input-ids) Marian uses the `pad_token_id` as the starting token for `decoder_input_ids` generation. If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see `past_key_values`). decoder_attention_mask (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*): Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also be used by default. Example: ```python >>> from transformers import AutoTokenizer, MarianModel >>> tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-de") >>> model = MarianModel.from_pretrained("Helsinki-NLP/opus-mt-en-de") >>> inputs = tokenizer("Studies have been shown that owning a dog is good for you", return_tensors="pt") >>> decoder_inputs = tokenizer( ... " Studien haben gezeigt dass es hilfreich ist einen Hund zu besitzen", ... return_tensors="pt", ... add_special_tokens=False, ... ) >>> outputs = model(input_ids=inputs.input_ids, decoder_input_ids=decoder_inputs.input_ids) >>> last_hidden_states = outputs.last_hidden_state >>> list(last_hidden_states.shape) [1, 26, 512] ```N)r#rsrrr!r@)r rr r#rsrrrrr)r rdecoder_hidden_statesdecoder_attentionsr/encoder_last_hidden_staterencoder_attentionsra) rJrrlenrKrr rrr r/) r<r#rsrrgrhrrrirrvdecoder_outputss r.r]MarianModel.forwardsh  ""ll#-+ OO_==-"1!"4474H14Loa0RV14_1E1I?1-tO,,  '1"1!"4#1+/    "-??+;;"1"?"?.99,==&5&G&G"1"?"?.99  r0)rIrKrJrH) NNNNNNNNN)rcrdrerf_keys_to_ignore_on_load_missingr"r;rOrTrXr[rhrrrerrrHr*rkrrr rJrrrrr]rlrmrns@r.rCrCps[ /.'# |*9.9*cbll0.2.2596:HL(,26:>!%V ##d*V  t+V !++d2 V !& t 3 V u||,>E V V ((4/V  %0047V $;V +,V  V V r0rCzX The Marian Model with a language modeling head. Can be used for summarization. ) custom_introcf^\rSrSrSr/SQrSS/rSS0rS\4U4S jjr S#S \ S \ S -S \ S\ R4U4SjjjrS$S \ S\ R4SjjrSrS \ SS 4SjrS\ R4Sjr\\S%S\R,S -S\R.S -S\R,S -S\R.S -S\\R.\-S -S\S -S\R6S -S\R6S -S\R,S -S\ S -S\\S\4S jj55rS\R.4S!jr S"r!U=r"$)&rir)rrDrErDrElm_head.weight!model.decoder.embed_tokens.weightrc>[TU]U5 [U5UlURR (a SSSS.UlUR (a URO URnURS[R"SU455 [R"URUSS9UlUR!5 g)Nzmodel.shared.weight)rvrwz!model.encoder.embed_tokens.weightrr!Fr)r:r;rCrrrGrIrr3register_bufferrHzerosrrrlm_headr)r<rtarget_vocab_sizer=s r.r;MarianMTModel.__init__+s   ( ;; 7 7"75J5J'D # 281X1XF--^d^w^w 0%++qBS>T2UVyy1BO  r0Nr]pad_to_multiple_of mean_resizingr7c>[TU]XU5nURR(aUR U5 U$rb)r:resize_token_embeddingsrrG_resize_final_logits_bias)r<r]r~rrcr=s r.r%MarianMTModel.resize_token_embeddings<s88]jk ;; 7 7  * *> :r0cUR5nURXAU5nURU5 URRSnUR R (aXR lUR R (a^UR5bMUR R(d2UR5nURXa5nURU5 UR5$)Nr) rOr`rTrCr)rrGr3get_output_embeddingstie_word_embeddings_get_resized_lm_headset_output_embeddings)r<r]r~argsrbrc old_lm_head new_lm_heads r._resize_token_embeddings&MarianMTModel._resize_token_embeddingsEs22455nVhi !!.1'..44Q7 ;; 7 7-;KK * KK 8 8**,8KK33446K33KPK  & &{ 3((**r0c:URR(a [S5eURR 5nUR X!5nURR U5 UR5bMURR(d2UR5nURXA5nURU5 URR 5nUcU$XRl UR5 URU5 U$r_)rrGr+rrXr`r[rrrrr3rar)r<r]rbrcrrrds r.re-MarianMTModel.resize_decoder_token_embeddings[s ;; 7 7K  @@B55nU //?  % % ' 3DKK<[<[446K33KPK  & &{ 3zz>>@  ! *8 &  &&~6r0c,URRSnX::aURSS2SU24nON[R"SX- 4URRS9n[R "URU/SS9nUR SU5 g)Nr'r!rryr)rr)rHrzrZcatry)r<r]old_num_tokensnew_bias extra_biass r.r'MarianMTModel._resize_final_logits_bias{s//55b9  +--a..@AHa)H%IRVRhRhRoRopJyy$"8"8*!E1MH 0(;r0rccXlgrb)r{)r<rcs r.r#MarianMTModel.set_output_embeddingss% r0r#rsrrgrhrrrilabelsrrvc vU bXU (a[RS5 Sn Uc7Uc4[XRRURR 5nUR "U4UUUUUUUU S.U D6n URU S5UR-n SnU bF[5nU"U RSURR5U RS55n[UU U RU RU RU R U R"U R$U R&S9 $)u decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*): Indices of decoder input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are decoder input IDs?](../glossary#decoder-input-ids) Marian uses the `pad_token_id` as the starting token for `decoder_input_ids` generation. If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see `past_key_values`). decoder_attention_mask (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*): Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also be used by default. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should either be in `[0, ..., config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. Example: ```python >>> from transformers import AutoTokenizer, MarianMTModel >>> src = "fr" # source language >>> trg = "en" # target language >>> model_name = f"Helsinki-NLP/opus-mt-{src}-{trg}" >>> model = MarianMTModel.from_pretrained(model_name) >>> tokenizer = AutoTokenizer.from_pretrained(model_name) >>> sample_text = "où est l'arrêt de bus ?" >>> batch = tokenizer([sample_text], return_tensors="pt") >>> generated_ids = model.generate(**batch) >>> tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] "Where's the bus stop?" ``` NzJThe `use_cache` argument is changed to `False` since `labels` is provided.F)rsrrhrgrrrirrr') losslogitsrrlrmr/rnrro)rwarningr/rr$r%rr{rrrr3rrrlrmr/rnrro)r<r#rsrrgrhrrrirrrvoutputs lm_logitsmasked_lm_lossloss_fcts r.r]MarianMTModel.forwards<p  klI (-B-J$6KK44dkk6X6X%!'+jj  ' )/+#9+'"7 '  ' LL,t/E/EE   ')H%innR9W9W&XZ`ZeZefhZijN#33")"?"?&99$55&-&G&G")"?"?&99  r0cj[XRRURR5$rb)r/rr$r%)r<rs r.%prepare_decoder_input_ids_from_labels3MarianMTModel.prepare_decoder_input_ids_from_labelss#!&++*B*BDKKDfDfggr0)rIr{r)NTrb) NNNNNNNNNN)#rcrdrerfrrs_keys_to_ignore_on_saverIr"r;rhrrrrrrerrrrrHr*rkrrr rJrrrr]rrlrmrns@r.rrs  '# FGmn*,OP|$ae!7:TzY] +s+_a_k_k+,@<<<&BLL&.2.2596:HL(,26:>*.!%] ##d*]  t+] !++d2 ] !& t 3 ] u||,>E ] ] ((4/]  %0047]   4'] $;] +,]  ] ] ~hELLhhr0rc2^\rSrSrSrU4SjrSrSrU=r$)MarianDecoderWrapperiz This wrapper class is a helper class to correctly load pretrained checkpoints when the causal language model is used in combination with the [`EncoderDecoderModel`] framework. cd>[TU]U5 [U5UlUR 5 grb)r:r;r,rKrr<rr=s r.r;MarianDecoderWrapper.__init__s&  $V,  r0c&UR"U0UD6$rbrK)r<rrvs r.r]MarianDecoderWrapper.forwards||T,V,,r0r) rcrdrerfrgr;r]rlrmrns@r.rrs  --r0rcj^\rSrSrSS0rU4SjrSrSr\\ SS\ RS-S \ RS-S \ RS-S \ RS-S \S-S \ RS-S\ RS-S\S-S\\ R-S\\S\\-4Sjj55rSrU=r$)MarianForCausalLMirvrwc>SUlSUl[TU] U5 [ U5Ul[ R"URURSS9Ul UR5 g)NTFr) rr8r:r;rrrr hidden_sizerr{rrs r.r;MarianForCausalLM.__init__sX $)!  )&1 yy!3!3V5F5FUS  r0cBURRR$rbrrKrrPs r.rO&MarianForCausalLM.get_input_embeddings szz!!...r0c8XRRlgrbrrSs r.rT&MarianForCausalLM.set_input_embeddingss*/ 'r0Nr#rsrrrrrrlogits_to_keeprvr7c URR"SUUUUUUUS.U D6n U Sn [U [5(a [ U *S5OU n UR U SS2U SS245nSnUbaUR UR5n[5nU"URSURR5URS55n[UUU RU RU RU R S9$)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, ..., 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, ..., config.vocab_size]`. Example: ```python >>> from transformers import AutoTokenizer, MarianForCausalLM >>> tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-fr-en") >>> model = MarianForCausalLM.from_pretrained("Helsinki-NLP/opus-mt-fr-en") >>> assert model.config.is_decoder, f"{model.__class__} has to be configured as a decoder." >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits >>> expected_shape = [1, inputs.input_ids.shape[-1], model.config.vocab_size] >>> list(logits.shape) == expected_shape True ```rkrNr')rrrrr r/ra)rrKrrhslicer{torZrrrrrrrr r/)r<r#rsrrrrrrrrvrr slice_indicesrrrs r.r]MarianForCausalLM.forwardsL>BZZ=O=O > )"7#9+' >  >   8B>SV8W8W~ot4]k mA}a,?@A  YYv}}-F')HFKKDKK,B,BCV[[QS_UD0#33!//))$55   r0)r{r) NNNNNNNNr)rcrdrerfrIr;rOrTrrrHr*rkrJr rrhrrrrr]rlrmrns@r.rrs/= /0.2.2:>;?(,26*.!%-.A ##d*A  t+A  %0047 A !& 1 1D 8 A  A ((4/A   4'A $;A ell*A +,A  2 2A A r0r)rrCrr)Nr)Jrgrcollections.abcrnumpyrDrHrtorch.nnrrr activationsr cache_utilsr r r generationr masking_utilsr rmodeling_flash_attention_utilsrmodeling_layersrmodeling_outputsrrrrrmodeling_utilsrrprocessing_utilsrutilsrrrrr utils.genericrutils.output_capturingrr configuration_marianr" get_loggerrcrrkrhr/rr2Modulerrrrrrr r,rCrrr__all__rar0r.rsD $ %&!CC)JB9G&8E.   H % %,, c [^  -",,-R!% II% <<% % << % LL4' % T\ %%'(%:r)biir)l030hO3OdO