+ jRt^RIHt^RIt^RIHt^RIHtHtHt^RI H t ^RI H t ^RIHt^RIHtHtHtHtHtHt^R IHtHt^R IHt^R IHt^R IHtH t H!t!H"t"^R I#H$t$^RI%H&t&^RI'H(t(]"PR!]*4t+]PXt-!RR]P\4t/R9RRllt0!RR]P\4t1!RR]P\4t2!RR]P\4t3!RR]P\4t4!RR]P\4t5!RR]4t6!R R!]P\4t7!R"R#]P\4t8!R$R%]P\4t9!R&R']P\4t:!R(R)]P\4t;] !R*R+]44t<] !R,R-]<44t=] !R.R/]<44t>] !R0R17!R2R3]<44t?] !R4R17!R5R6]<44t@] !R7R8]<44tA.R:OtBR#);zPyTorch LayoutLM model.)CallableN)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)initialization)ACT2FN)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPoolingMaskedLMOutputQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)apply_chunking_to_forward)TransformersKwargsauto_docstringcan_return_tuplelogging)merge_with_config_defaults)capture_outputsLayoutLMConfigc@aa]tRt^0toRtV3RltRRltRtVtV;t #)LayoutLMEmbeddingszGConstruct the embeddings from word, position and token_type embeddings.c<\SV`4\P!VPVP VP R7Vn\P!VPVP 4Vn \P!VPVP 4Vn \P!VPVP 4Vn \P!VPVP 4Vn \P!VPVP 4Vn\P!VPVP 4Vn\#VP VP$R7Vn\P(!VP*4Vn\P.!\0P2!VP4P5R4RR7VnR#)) padding_idxepsF) persistentN)super__init__r Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingsmax_2d_position_embeddingsx_position_embeddingsy_position_embeddingsh_position_embeddingsw_position_embeddingstype_vocab_sizetoken_type_embeddingsLayoutLMLayerNormlayer_norm_eps LayerNormDropouthidden_dropout_probdropoutBuffertorcharangeexpand position_idsselfconfig __class__s&&o/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/layoutlm/modeling_layoutlm.pyr'LayoutLMEmbeddings.__init__3sT !||F,=,=v?Q?Q_e_r_rs#%<<0N0NPVPbPb#c %'\\&2S2SU[UgUg%h"%'\\&2S2SU[UgUg%h"%'\\&2S2SU[UgUg%h"%'\\&2S2SU[UgUg%h"%'\\&2H2H&J\J\%]"*6+=+=6CXCXYzz&"<"<= IIell63Q3Q&R&Y&YZa&botucVeVP4pMVP4RRpV^,pVe VPM VPpVfVPRRV13,pVf(\P!V\P VR7pVfVP V4pTp VPV4p VPVR,4p VPVR,4p VPVR,4p VPVR,4pTPTR,TR,, 4pTPTR,TR,, 4pTPT4pT T ,T ,T ,T ,T,T,T,T,pTPT4pTPT4pT# \dp\R4ThRp?ii;i) NNNNdtypedevicez:The `bbox`coordinate values should be within 0-1000 range.r%)rIrI)rIrIr$)rIrI)rIrI)sizerLr@r=zeroslongr,r.r0r1 IndexErrorr2r3r5r8r;)rB input_idsbboxtoken_type_idsr@ inputs_embeds input_shape seq_lengthrLwords_embeddingsr.left_position_embeddingsupper_position_embeddingsright_position_embeddingslower_position_embeddingser2r3r5 embeddingss&&&&&& rEforwardLayoutLMEmbeddings.forwardBs  #..*K',,.s3K ^ %.%:!! @T@T  ,,Q  ^ J ! "& '( ((  ( (  ( $  $$ $$ $ ^^J/ \\*- ) bYZ`a a bs=A G G4# G//G4) r8r;r2r.r@r5r3r,r0r1)NNNNN) __name__ __module__ __qualname____firstlineno____doc__r'ra__static_attributes____classdictcell__ __classcell__rD __classdict__s@@rErr0sQ v55rGrcV^8dQhR\PR\PR\PR\PR\PR,R\R\/#) rNmodulequerykeyvalueattention_maskNscalingr;)rModuler=Tensorfloat)formats"rE __annotate__rx{sg%% II% <<% % << % LL4' %  %%rGc \P!WP^^44V,pVe W,p\PP VR\P R7PVP4p\PPWVPR7p\P!W4p V P^^4P4p W3#)rN)dimrK)ptrainingr%) r=matmul transposer functionalsoftmaxfloat32torKr;r| contiguous) rnrorprqrrrsr;kwargs attn_weights attn_outputs &&&&&&&, rEeager_attention_forwardr{s<<}}Q':;gEL!#4 ==((2U]](SVVW\WbWbcL==((6??([L,,|3K''1-88:K  $$rGcHaa]tRt^toV3RltRV3RlRlltRtVtV;t#)LayoutLMSelfAttentionc\<\SV`4VPVP,^8wd9\ VR4'g'\ RVP RVP R24hWnVPVn\VPVP, 4VnVPVP,Vn \P!VPVP4Vn \P!VPVP4Vn \P!VPVP4Vn\P!VP 4VnVP VnVPR,VnR#)rMembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ()Ng)r&r'r*num_attention_headshasattr ValueErrorrCintattention_head_size all_head_sizerLinearrorprqr9attention_probs_dropout_probr;attention_dropoutrsrAs&&rEr'LayoutLMSelfAttention.__init__sD     : : :a ?PVXhHiHi#F$6$6#78 445Q8   #)#=#= #&v'9'9FS>STzz&"<"<= rGch<V^8dQhRS[PRS[PRS[P/#rNr input_tensorrr=ru)rwrls"rErxLayoutLMSelfOutput.__annotate__1U\\RWR^R^rGcxVPV4pVPV4pVPW,4pV#rrr;r8rBrrs&&&rEraLayoutLMSelfOutput.forward5 =1  ]3 }'CD rGr8rr;rrks@@rErr> rGrcHaa]tRt^toV3RltRV3RlRlltRtVtV;t#)LayoutLMAttentioncd<\SV`4\V4Vn\ V4VnR#r)r&r'rrBroutputrAs&&rEr'LayoutLMAttention.__init__s& )&1 (0 rGc<V^8dQhRS[PRS[PR,RS[S[,RS[P/#rr=rurrr)rwrls"rErxLayoutLMAttention.__annotate__sL  || ))D0 +,   rGc ZTpVP!V3RV/VBwrVPW4pV#rr)rBr)rBrrrrresidual_s&&&, rEraLayoutLMAttention.forwardsE !99  )    M< rG)rrBrrrks@@rErrs1   rGrcDaa]tRt^toV3RltV3RlRltRtVtV;t#)LayoutLMIntermediatec"<\SV`4\P!VPVP 4Vn\VP\4'd\VP,Vn R#VPVn R#r) r&r'rrr*intermediate_sizer isinstance hidden_actstrrintermediate_act_fnrAs&&rEr'LayoutLMIntermediate.__init__s` YYv1163K3KL f'' - -'-f.?.?'@D $'-'8'8D $rGcN<V^8dQhRS[PRS[P/#rNrrr)rwrls"rErx!LayoutLMIntermediate.__annotate__s#U\\ellrGcJVPV4pVPV4pV#rrrrBrs&&rEraLayoutLMIntermediate.forwards& =1 00? rGrrrks@@rErrs9rGrcDaa]tRt^toV3RltV3RlRltRtVtV;t#)LayoutLMOutputc0<\SV`4\P!VPVP 4Vn\P!VP VPR7Vn\P!VP4Vn R#r) r&r'rrrr*rr8r7r9r:r;rAs&&rEr'LayoutLMOutput.__init__s` YYv779K9KL f&8&8f>S>STzz&"<"<= rGch<V^8dQhRS[PRS[PRS[P/#rr)rwrls"rErxLayoutLMOutput.__annotate__rrGcxVPV4pVPV4pVPW,4pV#rrrs&&&rEraLayoutLMOutput.forwardrrGrrrks@@rErrrrGrcNaa]tRtRtoV3RltRV3RlRlltRtRtVtV;t #) LayoutLMLayeric<\SV`4VPVn^Vn\ V4Vn\ V4Vn\V4Vn R#)r$N) r&r'chunk_size_feed_forward seq_len_dimr attentionr intermediaterrrAs&&rEr'LayoutLMLayer.__init__sI '-'E'E$*6208$V, rGc<V^8dQhRS[PRS[PR,RS[S[,RS[P/#rr)rwrls"rErxLayoutLMLayer.__annotate__sL||))D0+,   rGc VP!V3RV/VBp\VPVPVPV4pV#r)rrfeed_forward_chunkrr)rBrrrrs&&&,rEraLayoutLMLayer.forwardsW   )  2  # #T%A%A4CSCSUb rGcJVPV4pVPW!4pV#r)rr)rBattention_outputintermediate_output layer_outputs&& rEr LayoutLMLayer.feed_forward_chunk)s)"//0@A{{#6I rG)rrrrrr) rcrdrerfr'rarrhrirjrks@@rErrs-$rGrcHaa]tRtRtoV3RltRV3RlRlltRtVtV;t#)LayoutLMEncoderi0c<\SV`4Wn\P!\ VP 4Uu.uFp\V4NK up4VnRVn R#uupi)FN) r&r'rCr ModuleListrangenum_hidden_layersrlayergradient_checkpointing)rBrCirDs&& rEr'LayoutLMEncoder.__init__1sR  ]]5IaIaCb#cCbaM&$9Cb#cd &+#$dsA&cx<V^8dQhRS[PRS[PR,RS[S[,RS[/#r)r=rurrrr )rwrls"rErxLayoutLMEncoder.__annotate__7sH  || ))D0 +,    rGc TVPF pV!VV3/VBpK \VR7#))last_hidden_state)rr )rBrrrr layer_modules&&&, rEraLayoutLMEncoder.forward7s> !JJL(M'+  rG)rCrrrrrks@@rErr0s,    rGrcDaa]tRtRtoV3RltV3RlRltRtVtV;t#)LayoutLMPooleriJc<\SV`4\P!VPVP4Vn\P !4VnR#r)r&r'rrr*rTanh activationrAs&&rEr'LayoutLMPooler.__init__Ks9 YYv1163E3EF '')rGcN<V^8dQhRS[PRS[P/#rr)rwrls"rErxLayoutLMPooler.__annotate__Ps#U\\ellrGc\VR,pVPV4pVPV4pV#)rI)rIrM)rr )rBrfirst_token_tensor pooled_outputs&& rEraLayoutLMPooler.forwardPs2+40 #56  6 rG)r rrrks@@rErrJs$ rGrcDaa]tRtRtoV3RltV3RlRltRtVtV;t#)LayoutLMPredictionHeadTransformiZc<\SV`4\P!VPVP4Vn\ VP\4'd\VP,Vn MVPVn \P!VPVPR7Vn R#r) r&r'rrr*rrrrrtransform_act_fnr8r7rAs&&rEr'(LayoutLMPredictionHeadTransform.__init__[s~ YYv1163E3EF f'' - -$*6+<+<$=D !$*$5$5D !f&8&8f>S>STrGcN<V^8dQhRS[PRS[P/#rr)rwrls"rErx,LayoutLMPredictionHeadTransform.__annotate__ds#U\\ellrGclVPV4pVPV4pVPV4pV#r)rrr8rs&&rEra'LayoutLMPredictionHeadTransform.forwardds4 =1 --m< }5 rG)r8rrrrks@@rErrZsUrGrc8aa]tRtRtoV3RltRtRtVtV;t#)LayoutLMLMPredictionHeadilc<\SV`4\V4Vn\P !VP VPRR7Vn\P!\P!VP44Vn R#)T)biasN) r&r'r transformrrr*r)decoder Parameterr=rQrrAs&&rEr'!LayoutLMLMPredictionHead.__init__ms[ 8@yy!3!3V5F5FTR LLV->->!?@ rGcJVPV4pVPV4pV#r)r r!rs&&rEra LayoutLMLMPredictionHead.forwardvs$}5  ]3 rG)rr!r rrks@@rErrlsArGrcDaa]tRtRtoV3RltV3RlRltRtVtV;t#)LayoutLMOnlyMLMHeadi}cD<\SV`4\V4VnR#r)r&r'r predictionsrAs&&rEr'LayoutLMOnlyMLMHead.__init__~s 3F;rGcN<V^8dQhRS[PRS[P/#)rNsequence_outputrr)rwrls"rErx LayoutLMOnlyMLMHead.__annotate__s#!!u||! !rGc(VPV4pV#rr))rBr,prediction_scoress&& rEraLayoutLMOnlyMLMHead.forwards ,,_=  rGr/rrks@@rEr'r'}s<!!!rGr'czaa]tRtRtoRtRtR]R]/t] P!4V3Rl4t V3Rlt Rt VtV;t#) LayoutLMPreTrainedModelilayoutlmTr attentionsc <\SV`V4\V\4'd#\P !VP 4R#\V\4'db\P!VP\P!VPPR,4PR44R#R#)zInitialize the weightsNr%r#)r& _init_weightsrrinitzeros_rrcopy_r@r=r>rr?)rBrnrDs&&rEr7%LayoutLMPreTrainedModel._init_weightss~ f% f6 7 7 KK $  2 3 3 JJv**ELL9L9L9R9RSU9V,W,^,^_f,g h4rGc&<V^8dQh/S[;R&#)rNrCr)rwrls"rErx$LayoutLMPreTrainedModel.__annotate__s rG)rcrdrerfbase_model_prefixsupports_gradient_checkpointingrr_can_record_outputsr=no_gradr7__annotate_func__rhrirjrks@@rEr3r3sI#&*#+  ]]_iirGr3craa]tRtRtoV3RltRtRt]]] RV3RlRll444t Rt Vt V;t #) LayoutLMModelic<\SV`V4Wn\V4Vn\ V4Vn\V4VnVP4R#r) r&r'rCrr`rencoderrpooler post_initrAs&&rEr'LayoutLMModel.__init__sE   ,V4&v. $V,  rGc.VPP#rr`r,rBs&rEget_input_embeddings"LayoutLMModel.get_input_embeddingss...rGc&WPnR#rrL)rBrqs&&rEset_input_embeddings"LayoutLMModel.set_input_embeddingss */'rGc6<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[S[,R S[S[,/#) rNrTNrUrrrVr@rWrr)r= LongTensorrrrrr )rwrls"rErxLayoutLMModel.__annotate__s[ [ ##d*[ %[ ))D0 [ ((4/ [ &&- [ ((4/[ +,[  + +[ rGc lVeVe \R4hVe#VPW4VP4pM#VeVP4RR pM \R4hVe VPM VPp Vf\P !WR7pVf(\P !V\PV R7pVf/\P !VR ,\PV R7pVP^4P^4p V PVPR7p RV , \P!VP4P,p VPVVVVVR7p VP!V V 3/VBp V ^,p VPV 4p\!V VR 7#) a bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. See [Overview](#Overview) for normalization. Examples: ```python >>> from transformers import AutoTokenizer, LayoutLMModel >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = LayoutLMModel.from_pretrained("microsoft/layoutlm-base-uncased") >>> words = ["Hello", "world"] >>> normalized_word_boxes = [637, 773, 693, 782], [698, 773, 733, 782] >>> token_boxes = [] >>> for word, box in zip(words, normalized_word_boxes): ... word_tokens = tokenizer.tokenize(word) ... token_boxes.extend([box] * len(word_tokens)) >>> # add bounding boxes of cls + sep tokens >>> token_boxes = [[0, 0, 0, 0]] + token_boxes + [[1000, 1000, 1000, 1000]] >>> encoding = tokenizer(" ".join(words), return_tensors="pt") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = torch.tensor([token_boxes]) >>> outputs = model( ... input_ids=input_ids, bbox=bbox, attention_mask=attention_mask, token_type_ids=token_type_ids ... ) >>> last_hidden_states = outputs.last_hidden_state ```NzDYou cannot specify both input_ids and inputs_embeds at the same timez5You have to specify either input_ids or inputs_embeds)rLrJ)rKg?)rTrUr@rVrW)r pooler_outputr%))r%warn_if_padding_and_no_attention_maskrPrLr=onesrQrR unsqueezerrKfinfominr`rGrHr )rBrTrUrrrVr@rWrrXrLextended_attention_maskembedding_outputencoder_outputsr,rs&&&&&&&, rEraLayoutLMModel.forwardsf  ]%>cd d  "  6 6y Q#..*K  &',,.s3KTU U%.%:!! @T@T  !"ZZ CN  !"[[EJJvVN <;;{T1FSD"0":":1"="G"G"J"9"<"<4::"<"N#&)@#@EKKPTPZPZD[D_D_"_??%)' + ,,  #   *!, O4 )-'  rG)rCr`rGrH)NNNNNN)rcrdrerfr'rNrQrrrrarhrirjrks@@rErErEsF /0 [ [  [ [ rGrEczaa]tRtRtoRRRR/tV3RltRtRtR t] ] R V3R lR ll44t R t Vt V;t#)LayoutLMForMaskedLMizcls.predictions.decoder.biaszcls.predictions.biaszcls.predictions.decoder.weightz*layoutlm.embeddings.word_embeddings.weightc<\SV`V4\V4Vn\ V4VnVP 4R#r)r&r'rEr4r'clsrIrAs&&rEr'LayoutLMForMaskedLM.__init__s4  %f- &v. rGcBVPPP#rr4r`r,rMs&rErN(LayoutLMForMaskedLM.get_input_embeddings}}''777rGcBVPPP#r)rer)r!rMs&rEget_output_embeddings)LayoutLMForMaskedLM.get_output_embeddings"sxx##+++rGcWPPnVPVPPnR#r)rer)r!r)rBnew_embeddingss&&rEset_output_embeddings)LayoutLMForMaskedLM.set_output_embeddings%s*'5$$2$7$7!rGc^<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[S[,R S[S[,/ # rNrTNrUrrrVr@rWlabelsrr)r=rTrrrrr )rwrls"rErx LayoutLMForMaskedLM.__annotate__+sU U ##d*U %U ))D0 U ((4/ U &&- U ((4/U   4'U +,U   U rGc LVP!VV3RVRVRVRV/VBp V ^,p VPV 4p Rp VeG\4p V !V PRVPP 4VPR44p \ V V V PV PR7#)a bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. See [Overview](#Overview) for normalization. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.vocab_size]` (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]` Examples: ```python >>> from transformers import AutoTokenizer, LayoutLMForMaskedLM >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = LayoutLMForMaskedLM.from_pretrained("microsoft/layoutlm-base-uncased") >>> words = ["Hello", "[MASK]"] >>> normalized_word_boxes = [637, 773, 693, 782], [698, 773, 733, 782] >>> token_boxes = [] >>> for word, box in zip(words, normalized_word_boxes): ... word_tokens = tokenizer.tokenize(word) ... token_boxes.extend([box] * len(word_tokens)) >>> # add bounding boxes of cls + sep tokens >>> token_boxes = [[0, 0, 0, 0]] + token_boxes + [[1000, 1000, 1000, 1000]] >>> encoding = tokenizer(" ".join(words), return_tensors="pt") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = torch.tensor([token_boxes]) >>> labels = tokenizer("Hello world", return_tensors="pt")["input_ids"] >>> outputs = model( ... input_ids=input_ids, ... bbox=bbox, ... attention_mask=attention_mask, ... token_type_ids=token_type_ids, ... labels=labels, ... ) >>> loss = outputs.loss ```rrrVr@rWNlosslogitsrr5r%) r4rerrrCr)r rr5)rBrTrUrrrVr@rWrtroutputsr,r0masked_lm_lossloss_fcts&&&&&&&&, rEraLayoutLMForMaskedLM.forward)sz--   * *  &  (   "!* HH_5  ')H%!&&r4;;+A+AB BN $!//))   rG)rer4NNNNNNN)rcrdrerf_tied_weights_keysr'rNrlrprrrarhrirjrks@@rErcrcsY '(>(*V 8,8U U U U rGrcz LayoutLM Model with a sequence classification head on top (a linear layer on top of the pooled output) e.g. for document image classification tasks such as the [RVL-CDIP](https://www.cs.cmu.edu/~aharley/rvl-cdip/) dataset. ) custom_introcbaa]tRtRtoV3RltRt]]RV3RlRll44tRt Vt V;t #)!LayoutLMForSequenceClassificationic2<\SV`V4VPVn\V4Vn\ P !VP4Vn\ P!VPVP4Vn VP4R#r r&r' num_labelsrEr4rr9r:r;rr* classifierrIrAs&&rEr'*LayoutLMForSequenceClassification.__init__i   ++%f- zz&"<"<= ))F$6$68I8IJ rGcBVPPP#rrhrMs&rErN6LayoutLMForSequenceClassification.get_input_embeddingsrjrGc^<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[S[,R S[S[,/ #rs)r=rTrrrrr)rwrls"rErx.LayoutLMForSequenceClassification.__annotate__sf f ##d*f %f ))D0 f ((4/ f &&- f ((4/f   4'f +,f  ) )f rGc VP!R RVRVRVRVRVRV/VBp V ^,p VPV 4p VPV 4p Rp VEeVPPfVP ^8XdRVPnMrVP ^8dQVP \P8Xg VP \P8XdR VPnMR VPnVPPR8XdK\4p VP ^8Xd'V !V P4VP44p MV !W4p MVPPR 8Xd>\4p V !V PR VP 4VPR 44p M-VPPR 8Xd\4p V !W4p \V V V P V P"R 7#)a  bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. See [Overview](#Overview) for normalization. labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). Examples: ```python >>> from transformers import AutoTokenizer, LayoutLMForSequenceClassification >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = LayoutLMForSequenceClassification.from_pretrained("microsoft/layoutlm-base-uncased") >>> words = ["Hello", "world"] >>> normalized_word_boxes = [637, 773, 693, 782], [698, 773, 733, 782] >>> token_boxes = [] >>> for word, box in zip(words, normalized_word_boxes): ... word_tokens = tokenizer.tokenize(word) ... token_boxes.extend([box] * len(word_tokens)) >>> # add bounding boxes of cls + sep tokens >>> token_boxes = [[0, 0, 0, 0]] + token_boxes + [[1000, 1000, 1000, 1000]] >>> encoding = tokenizer(" ".join(words), return_tensors="pt") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = torch.tensor([token_boxes]) >>> sequence_label = torch.tensor([1]) >>> outputs = model( ... input_ids=input_ids, ... bbox=bbox, ... attention_mask=attention_mask, ... token_type_ids=token_type_ids, ... labels=sequence_label, ... ) >>> loss = outputs.loss >>> logits = outputs.logits ```rTrUrrrVr@rWN regressionsingle_label_classificationmulti_label_classificationrwr>r%)r4r;rrC problem_typerrKr=rRrrsqueezerrrrrr5)rBrTrUrrrVr@rWrtrrzrryrxr|s&&&&&&&&, rEra)LayoutLMForSequenceClassification.forwardsz--   * *  &  (      ]3 /  {{''/??a'/;DKK,__q(fllejj.HFLL\a\e\eLe/LDKK,/KDKK,{{''<7"9??a'#FNN$4fnn6FGD#F3D))-JJ+- B @&++b/R))-II,./'!//))   rGrr;r4rr~ rcrdrerfr'rNrrrarhrirjrks@@rErrs98f f f f rGra3 LayoutLM Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for sequence labeling (information extraction) tasks such as the [FUNSD](https://guillaumejaume.github.io/FUNSD/) dataset and the [SROIE](https://rrc.cvc.uab.es/?ch=13) dataset. cbaa]tRtRtoV3RltRt]]RV3RlRll44tRt Vt V;t #)LayoutLMForTokenClassificationic2<\SV`V4VPVn\V4Vn\ P !VP4Vn\ P!VPVP4Vn VP4R#rrrAs&&rEr''LayoutLMForTokenClassification.__init__ rrGcBVPPP#rrhrMs&rErN3LayoutLMForTokenClassification.get_input_embeddingsrjrGc^<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[S[,R S[S[,/ #rs)r=rTrrrrr)rwrls"rErx+LayoutLMForTokenClassification.__annotate__sR R ##d*R %R ))D0 R ((4/ R &&- R ((4/R   4'R +,R  & &R rGc ^VP!R RVRVRVRVRVRV/VBp V ^,p VPV 4p VPV 4p Rp Ve=\4p V !V P R VP 4VP R 44p \ V V V PV PR7#) a bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. See [Overview](#Overview) for normalization. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`. Examples: ```python >>> from transformers import AutoTokenizer, LayoutLMForTokenClassification >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = LayoutLMForTokenClassification.from_pretrained("microsoft/layoutlm-base-uncased") >>> words = ["Hello", "world"] >>> normalized_word_boxes = [637, 773, 693, 782], [698, 773, 733, 782] >>> token_boxes = [] >>> for word, box in zip(words, normalized_word_boxes): ... word_tokens = tokenizer.tokenize(word) ... token_boxes.extend([box] * len(word_tokens)) >>> # add bounding boxes of cls + sep tokens >>> token_boxes = [[0, 0, 0, 0]] + token_boxes + [[1000, 1000, 1000, 1000]] >>> encoding = tokenizer(" ".join(words), return_tensors="pt") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = torch.tensor([token_boxes]) >>> token_labels = torch.tensor([1, 1, 0, 0]).unsqueeze(0) # batch size of 1 >>> outputs = model( ... input_ids=input_ids, ... bbox=bbox, ... attention_mask=attention_mask, ... token_type_ids=token_type_ids, ... labels=token_labels, ... ) >>> loss = outputs.loss >>> logits = outputs.logits ```rTrUrrrVr@rWNrwr>r%) r4r;rrrrrrr5)rBrTrUrrrVr@rWrtrrzr,ryrxr|s&&&&&&&&, rEra&LayoutLMForTokenClassification.forwardsv--   * *  &  (   "!*,,71  ')HFKKDOO>> from transformers import AutoTokenizer, LayoutLMForQuestionAnswering >>> from datasets import load_dataset >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("impira/layoutlm-document-qa", add_prefix_space=True) >>> model = LayoutLMForQuestionAnswering.from_pretrained("impira/layoutlm-document-qa", revision="1e3ebac") >>> dataset = load_dataset("nielsr/funsd", split="train") >>> example = dataset[0] >>> question = "what's his name?" >>> words = example["words"] >>> boxes = example["bboxes"] >>> encoding = tokenizer( ... question.split(), words, is_split_into_words=True, return_token_type_ids=True, return_tensors="pt" ... ) >>> bbox = [] >>> for i, s, w in zip(encoding.input_ids[0], encoding.sequence_ids(0), encoding.word_ids(0)): ... if s == 1: ... bbox.append(boxes[w]) ... elif i == tokenizer.sep_token_id: ... bbox.append([1000] * 4) ... else: ... bbox.append([0] * 4) >>> encoding["bbox"] = torch.tensor([bbox]) >>> word_ids = encoding.word_ids(0) >>> outputs = model(**encoding) >>> loss = outputs.loss >>> start_scores = outputs.start_logits >>> end_scores = outputs.end_logits >>> start, end = word_ids[start_scores.argmax(-1)], word_ids[end_scores.argmax(-1)] >>> print(" ".join(words[start : end + 1])) M. Hamann P. Harper, P. Martinez ```rTrUrrrVr@rW)rzN) ignore_index)rx start_logits end_logitsrr5r>r%) r4rsplitrrlenrPclamprr rr5)rBrTrUrrrVr@rWrrrrzr,ryrr total_loss ignored_indexr| start_lossend_losss&&&&&&&&&, rEra$LayoutLMForQuestionAnswering.forwards~--   * *  &  (   "!*1#)<<r<#: #++B/::< ''+668   &=+D?'')*Q."1"9"9""==%%'(1, - 5 5b 9 (--a0M-33A}EO)//=AM']CH!,@J :H$x/14J+%!!//))   rG)r4rr)T)NNNNNNNNrrks@@rErrns9 8e e e e rGr)r)rcrrrrEr3)Crgcollections.abcrr=rtorch.nnrrrrr8 activationsrmodeling_layersr modeling_outputsr r r r rrmodeling_utilsrrprocessing_utilsr pytorch_utilsrutilsrrrr utils.genericrutils.output_capturingrconfiguration_layoutlmr get_loggerrcloggerr8r6rtrrrrrrrrrrrrr'r3rErcrrr__all__r>rGrErs$ AA&!9G&6RR752   H %LLGGV%.3)BII3)n   . 299  RYY .D bii 4 RYY  bii$ ryy "!"))!ioii&p +p p fp 1p p f v (?v  v rb %<b b Jy #:y y x rG