+ j"Rt^RIt^RIt^RIHt^RIHtHtHt^RIH t ^RI H t ^RI Ht^RIHtHtHtHtHt^RIHt^R IHt^R IHt^R IHtHtHtHt^R I H!t!H"t"H#t#^R I$H%t%^RI&H't']!4'dN^RI(t(^RI)H*t*]PPVPX](PZP\P^n,]P`!]14t2!RR]PV4t3!RR]PV4t4!RR]PV4t5!RR]PV4t6!RR]PV4t7!RR]PV4t8!RR]4t9R3Rlt:!RR ]PV4t;]!R!R"]44t!R&R']PV4t?]!R(R)]<44t@]!R*R+7!R,R-]<44tA]!R.R+7!R/R0]<44tB]!R1R2]<44tC.R5OtDR#)6zPyTorch LayoutLMv2 model.N)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)initialization)ACT2FN)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPoolingQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)PreTrainedModel)Unpack)apply_chunking_to_forward)auto_docstringis_detectron2_availableloggingrequires_backends)TransformersKwargscan_return_tuplemerge_with_config_defaults)capture_outputsLayoutLMv2Config)META_ARCH_REGISTRYc<aa]tRt^5toRtV3RltRtRtVtV;t #)LayoutLMv2EmbeddingszGConstruct the embeddings from word, position and token_type embeddings.c<\SV`4\P!VPVP VP R7Vn\P!VPVP 4Vn \P!VPVP4Vn \P!VPVP4Vn \P!VPVP4Vn\P!VPVP4Vn\P!VP"VP 4Vn\P&!VP VP(R7Vn\P*!VP,4Vn\P0!\2P4!VP4P7R4RR7VnR#)) padding_idxepsF persistentN)super__init__r Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingsmax_2d_position_embeddingscoordinate_sizex_position_embeddingsy_position_embeddings shape_sizeh_position_embeddingsw_position_embeddingstype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutBuffertorcharangeexpand position_idsselfconfig __class__s&&s/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/layoutlmv2/modeling_layoutlmv2.pyr(LayoutLMv2Embeddings.__init__8sX !||F,=,=v?Q?Q_e_r_rs#%<<0N0NPVPbPb#c %'\\&2S2SU[UkUk%l"%'\\&2S2SU[UkUk%l"%'\\&2S2SU[UfUf%g"%'\\&2S2SU[UfUf%g"%'\\&2H2H&J\J\%]"f&8&8f>S>STzz&"<"<= IIell63Q3Q&R&Y&YZa&botucVPVR,4pVPVR,4pVPVR,4pVPVR,4pTPTR,TR,, 4pTP TR,TR,, 4p\ P !TTTTTT.RR7p T # \dp\R4ThRp?ii;i) NNNz;The `bbox` coordinate values should be within 0-1000 range.NdimrKrK)rKrKr%)rKrKrKrKr&)r2r3 IndexErrorr5r6r?cat) rDbboxleft_position_embeddingsupper_position_embeddingsright_position_embeddingslower_position_embeddingser5r6spatial_position_embeddingss && rG!_calc_spatial_position_embeddings6LayoutLMv2Embeddings._calc_spatial_position_embeddingsHs c'+'A'A$w-'P $(,(B(B4=(Q %(,(B(B4=(Q %(,(B(B4=(Q %!% : :4=4PW=;X Y $ : :4=4PW=;X Y&+ii()))%%   ' #+*# cZ[ab b csA C C* C%%C*) r9r=r5r/rBr8r6r-r2r3) __name__ __module__ __qualname____firstlineno____doc__r(r\__static_attributes____classdictcell__ __classcell__rF __classdict__s@@rGrr5sQv ++rIrcNaa]tRt^btoV3RltRtRV3RlRlltRtVtV;t #)LayoutLMv2SelfAttentionc<\SV`4VPVP,^8wd9\ VR4'g'\ RVP RVP R24hVP VnVPVn\VPVP, 4VnVPVP,Vn VPVn VPVn VP 'd\P!VP^VP,RR7Vn\P!\ P"!^^VP44Vn\P!\ P"!^^VP44VnM\P!VPVP4Vn\P!VPVP4Vn\P!VPVP4Vn\P.!VP04VnR#)rOembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ()FbiasN)r'r(r+num_attention_headshasattr ValueErrorfast_qkvintattention_head_size all_head_sizehas_relative_attention_biashas_spatial_attention_biasrLinear qkv_linear Parameterr?zerosq_biasv_biasquerykeyvaluer;attention_probs_dropout_probr=rCs&&rGr( LayoutLMv2SelfAttention.__init__cs     : : :a ?PVXhHiHi#F$6$6#78 445Q8  #)#=#= #&v'9'9FS"S%**,CD --rI) rurtr=rrrvrwrror|ryr~r}rNNN) r^r_r`rar(rrrcrdrerfs@@rGriribsG4"$.$.$.rIricHaa]tRt^toV3RltRV3RlRlltRtVtV;t#)LayoutLMv2Attentioncd<\SV`4\V4Vn\ V4VnR#N)r'r(rirDLayoutLMv2SelfOutputoutputrCs&&rGr(LayoutLMv2Attention.__init__s& +F3 *62 rIc0<V^8dQhRS[S[,/#rr)rrgs"rGr LayoutLMv2Attention.__annotate__s  +,  rIc `TpVP!VV3RVRV/VBwrxVPWv4pV#rr)rDr) rDrrrrrresidualattention_output_s &&&&&, rGrLayoutLMv2Attention.forwardsW!"ii    "     ;;'7BrI)rrDr r^r_r`rar(rrcrdrerfs@@rGrrs3    rIrc8aa]tRt^toV3RltRtRtVtV;t#)rc0<\SV`4\P!VPVP4Vn\P !VPVPR7Vn\P!VP4Vn R#r N) r'r(rrxr+denser9r:r;r<r=rCs&&rGr(LayoutLMv2SelfOutput.__init__s` YYv1163E3EF f&8&8f>S>STzz&"<"<= rIcxVPV4pVPV4pVPW,4pV#rrr=r9rDr input_tensors&&&rGrLayoutLMv2SelfOutput.forward5 =1  ]3 }'CD rIr9rr=rrfs@@rGrrs> rIrcDaa]tRt^toV3RltV3RlRltRtVtV;t#)LayoutLMv2Intermediatec"<\SV`4\P!VPVP 4Vn\VP\4'd\VP,Vn R#VPVn R#r) r'r(rrxr+intermediate_sizer isinstance hidden_actstrrintermediate_act_fnrCs&&rGr(LayoutLMv2Intermediate.__init__s` YYv1163K3KL f'' - -'-f.?.?'@D $'-'8'8D $rIcN<V^8dQhRS[PRS[P/#)rPrreturnr?Tensor)rrgs"rGr#LayoutLMv2Intermediate.__annotate__s#U\\ellrIcJVPV4pVPV4pV#rrr)rDrs&&rGrLayoutLMv2Intermediate.forwards& =1 00? rIrrrfs@@rGrrs9rIrcDaa]tRt^toV3RltV3RlRltRtVtV;t#)LayoutLMv2Outputc0<\SV`4\P!VPVP 4Vn\P!VP VPR7Vn\P!VP4Vn R#r) r'r(rrxrr+rr9r:r;r<r=rCs&&rGr(LayoutLMv2Output.__init__s` YYv779K9KL f&8&8f>S>STzz&"<"<= rIch<V^8dQhRS[PRS[PRS[P/#)rPrrrr)rrgs"rGrLayoutLMv2Output.__annotate__s1U\\RWR^R^rIcxVPV4pVPV4pVPW,4pV#rrrs&&&rGrLayoutLMv2Output.forwardrrIrrrfs@@rGrrs> rIrcNaa]tRt^toV3RltRV3RlRlltRtRtVtV;t #)LayoutLMv2Layerc<\SV`4VPVn^Vn\ V4Vn\ V4Vn\V4Vn R#)r%N) r'r(chunk_size_feed_forward seq_len_dimr attentionr intermediaterrrCs&&rGr(LayoutLMv2Layer.__init__sI '-'E'E$,V426:&v. rIc0<V^8dQhRS[S[,/#rr)rrgs"rGrLayoutLMv2Layer.__annotate__s+,rIc VPVVVVR7p\VPVPVPV4pV#)r)rrfeed_forward_chunkrr) rDrroutput_attentionsrrrr layer_outputs &&&&&&, rGrLayoutLMv2Layer.forwardsT >>  ! * 1  # #T%A%A4CSCSUe rIcJVPV4pVPW!4pV#r)rr)rDrintermediate_outputrs&& rGr"LayoutLMv2Layer.feed_forward_chunks)"//0@A{{#6I rI)rrrrr)NFNN) r^r_r`rar(rrrcrdrerfs@@rGrrs/,rIrc^pV'dAV^,pW@^8P4V,, p\P!V4pM,\P!V)\P!V44pV^,pWV8pV\P !VP 4V, 4\P !W6, 4, W&, ,P\P4,p\P!V\P!W^, 44pV\P!WuV4, pV#)aR Adapted from Mesh Tensorflow: https://github.com/tensorflow/mesh/blob/0cb87fe07da627bf0b7e60475d59f95ed6b5be3d/mesh_tensorflow/transformer/transformer_layers.py#L593 Translate relative position to a bucket number for relative attention. The relative position is defined as memory_position - query_position, i.e. the distance in tokens from the attending position to the attended-to position. If bidirectional=False, then positive relative positions are invalid. We use smaller buckets for small absolute relative_position and larger buckets for larger absolute relative_positions. All relative positions >=max_distance map to the same bucket. All relative positions <=-max_distance map to the same bucket. This should allow for more graceful generalization to longer sequences than the model has been trained on. Args: relative_position: an int32 Tensor bidirectional: a boolean - whether the attention is bidirectional num_buckets: an integer max_distance: an integer Returns: a Tensor with the same shape as relative_position, containing int32 values in the range [0, num_buckets) ) longr?absmax zeros_likelogrrrr full_likewhere) relative_position bidirectional num_buckets max_distanceretn max_exactis_small val_if_larges &&&& rGrelative_position_bucketr!s* C  A%++- ;; II' ( II((%*:*:;L*M Nq I}H !'')i'(488L4L+MMQ\QhibnL99\5??B?? C cVR,pVR,pVPR4VPR4, pVPR4VPR4, p\VVPVPR7p\VVPVPR7p\P !4;_uu_4VP PP4V,P^^^^4pVPPP4V,P^^^^4pRRR4VP4pVP4pWg,pV# +'giL:;i)rKr#NrNrQrr&) r$rrrr?r%rr&r'rrr) rDrUposition_coord_xposition_coord_yrel_pos_x_2d_matrel_pos_y_2d_mat rel_pos_x rel_pos_yrs && rG!_calculate_2d_position_embeddings3LayoutLMv2Encoder._calculate_2d_position_embeddingsssF==+55b9R7R#R#\V\@PBPD4'd\ PF!VPH4\ P!VPJ4\ P!VPL4\ PN!VPPRVPR, 4R#R#)zInitialize the weightsvisual_segment_embeddingg)meanstd?Nr&r$)*r' _init_weightsrrirErrinitzeros_r|r}rcopy_rBr?r@rrALayoutLMv2VisualBackbonelencfgMODEL PIXEL_MEAN pixel_meanrr pixel_std PIXEL_STDLayoutLMv2Modelrpnormal_r>initializer_range detectron2layersFrozenBatchNorm2dones_r&rn running_mean constant_ running_varr!)rDmodule num_channelsrFs&& rGrB'LayoutLMv2PreTrainedModel._init_weightss f% f5 6 6{{### FMM* FMM*$ 4 5 5 JJv**ELL9L9L9R9RSU9V,W,^,^_f,g h  8 9 9vzz//::;L JJv((%,,vzz7G7G7R7R*S*X*XYeghjk*l m JJv''fjj6F6F6P6P)Q)V)VWcefhi)j k  0 0v9:: V<<3DKKLiLij; 1 1 C C D D JJv}} % KK $ KK++ , NN6--sVZZ/? @ ErIc&<V^8dQh/S[;R&#)rPrEr)rrgs"rGr&LayoutLMv2PreTrainedModel.__annotate__s rI)imagetext) r^r_r`rabase_model_prefixinput_modalitiesr?r%rB__annotate_func__rcrdrerfs@@rGr;r;s4%( ]]_AArIr;c\V\PPPP 4'd*\PP PW4#Tp\V\PP4'd\PP VPVPRRVR7p\PPVP4Vn\PPVP4VnVP VnVP"Vn\P$!R^\P&VP P(R7VnVP-4F wr4VP/V\1WA44K" ?V#)T) num_featuresr!affinetrack_running_stats process_grouprdevicer])rr?rmodules batchnorm _BatchNorm SyncBatchNormconvert_sync_batchnormrQrRrSrdr!rzr&rnrUrWfullrrinum_batches_trackednamed_children add_modulemy_convert_sync_batchnorm)rXrg module_outputnamechilds&& rGrsrss@&%((**44??@@zz''>>vUUM&*++==>>..,,  $' /  %xx11&--@ "XX// < %+%8%8 "$*$6$6 !,1JJr1EJJW]WjWjWqWq,r ),,.   '@'VW/ rIc>aa]tRtRtoV3RltRtRtRtVtV;t #)rFic <\SV`4VP4VnVPPP p\ P!V4!VP4p\VP\PPP4'gQhVPVn \VPPP4\VPPP48XgQh\VPPP4p\ P"!\$P&!VPPP4P)V^^4RR7Vn\ P"!\$P&!VPPP4P)V^^4RR7VnRVn\$P0!4'Ed\2P5R4RpVPP74VP.,P8p\ P:!\<P>!\<P>!V^,V, 4VP@^,, 4\<P>!\<P>!V^,V, 4VP@^,, 434Vn!M,\ PD!VP@R,4Vn!\VP@4^8XdOVP@PGVPP74VP.,PH4VPP74VP.,PHVP@^,8XgQhR#)r%Fr"p2z0using `AvgPool2d` instead of `AdaptiveAvgPool2d`:NrPNN)rz)%r'r(get_detectron2_configrHrIMETA_ARCHITECTURErgetrbackbonerQmodelingFPNrGrJrMrr>r?rrrKrLout_feature_key$are_deterministic_algorithms_enabledloggerwarning output_shapestride AvgPool2drceilimage_feature_pool_shapepoolAdaptiveAvgPool2dappendchannels)rDrE meta_archmodelrY input_shapebackbone_striderFs&& rGr(!LayoutLMv2VisualBackbone.__init__s //1HHNN44 "&&y1$((;%..**=*=*F*F*J*JKKKK 488>>,,-TXX^^5M5M1NNNN488>>445 ))ELL1J1J$K$P$PQ]_`bc$dqvw5<<0H0H#I#N#N|]^`a#botu#  5 5 7 7 NNM N$K"mm88:4;O;OPWWO IIdii A(HIFLkLklmLnnoIIdii A(HIFLkLklmLnnoDI,,V-L-LR-PQDI v.. /1 4  + + 2 24==3M3M3OPTPdPd3e3n3n o}}))+D,@,@AJJfNmNmnoNpppprIcb\P!V4'dTM VPVP, VP, pVP V4pW0P ,pVPV4P^R7P^^4P4pV#)rP) start_dim) r? is_tensortensorrKrLr~rrflattenrr)rDimages images_inputfeaturess&& rGr LayoutLMv2VisualBackbone.forwards#(??6#:#: QUQ`Q``dhdrdrr ==.00199X&...;EEaKVVXrIc T\PP4'dH\PP4'd$\PP 4R8g \ R4h\PP 4p\P P4p\PP4pW2,^8Xg \ R4h\W2,4Uu.uF,p\\WB,V^,V,44NK. pp\W2,4Uu.uF)p\PPWT,R7NK+ ppW,p\VPWg,R7Vn R#uupiuupi)r%z/Make sure torch.distributed is set up properly.zGMake sure the number of processes can be divided by the number of nodes)ranks)rgNr&)r? distributed is_availableis_initializedget_rank RuntimeErrorcuda device_countget_world_sizerlist new_grouprsr~)rD self_rank node_size world_sizeinode_global_rankssync_bn_groups node_ranks& rGsynchronize_batch_norm/LayoutLMv2VisualBackbone.synchronize_batch_norms@    * * , ,!!0022!!**,r1PQ Q%%..0 JJ++- &&557 &!+hi iV[\f\sVtuVtQRT% A7J"KLVtuMRS]SjMk MkE   ' '.?.B ' CMk  * 1$--~Ohi v s <2F /F%)r~rHrrKrLr) r^r_r`rar(rrrcrdrerfs@@rGrFrFsq:jjrIrFc8aa]tRtRtoV3RltRtRtVtV;t#)LayoutLMv2Pooleric<\SV`4\P!VPVP4Vn\P !4VnR#r)r'r(rrxr+rTanh activationrCs&&rGr(LayoutLMv2Pooler.__init__s9 YYv1163E3EF '')rIc\VR,pVPV4pVPV4pV#)rK)rKrO)rr)rDrfirst_token_tensor pooled_outputs&& rGrLayoutLMv2Pooler.forwards2+40 #56  6 rI)rrrrfs@@rGrrs$ rIrcaa]tRtRtoR]R]/tV3RltRtRt RRlt Rt R t RR lt ]]]RV3R lR ll444tR tVtV;t#)rNi&r attentionsc<\VR4\SV` V4WnVPVn\ V4Vn\V4Vn\P!VPR,VP4Vn VP'dL\P!\P!^VP4P ^,4Vn\P$!VPVP&R7Vn\P*!VP,4Vn\1V4Vn\5V4VnVP94R#)rQr Nr&)rr'r(rEhas_visual_segment_embeddingr embeddingsrFvisualrrxrr+ visual_projrzr)r&r>r9r:visual_LayerNormr;r<visual_dropoutrencoderrpooler post_initrCs&&rGr(LayoutLMv2Model.__init__*s$ -   ,2,O,O).v6.v6 99V%D%DR%H&J\J\]  , , ,,.LLaI[I[9\9c9cde9f,gD ) " V-?-?VEZEZ [ jj)C)CD(0 &v.  rIc.VPP#rrr-rDs&rGget_input_embeddings$LayoutLMv2Model.get_input_embeddings>s...rIc&WPnR#rr)rDrs&&rGset_input_embeddings$LayoutLMv2Model.set_input_embeddingsAs */'rIcVeVP4pMVP4RRpV^,pVfR\P!V\PVPR7pVP ^4P V4pVf\P!V4pVfVPPV4pVPPV4pVPPV4p VPPV4p WX,V ,V ,p VPPV 4p VPPV 4p V #)Nrhr&)rr?r@rrir$ expand_asrrr-r/r\r8r9r=) rD input_idsrUrBtoken_type_ids inputs_embedsr seq_lengthr/r[r8rs &&&&&& rG_calc_text_embeddings%LayoutLMv2Model._calc_text_embeddingsDs!  #..*K',,.s3K ^   << %**YM]M]^L'11!4>>yIL  !"--i8N   OO;;IFM"ooAA,O&*oo&W&WX\&]# $ E En U"8;VVYnn __..z: __,,Z8 rIc\VPVPV44pVPPV4pVPP V4pWE,V,pVP 'dWpP , pVPV4pVPV4pV#r) rrrr/r\rr>rr)rDr^rUrBvisual_embeddingsr/r[rs&&&& rG_calc_img_embeddings$LayoutLMv2Model._calc_img_embeddings]s ,,T[[-?@"ooAA,O&*oo&W&WX\&]#& rIcVeVe \R4hVeVP4#VeVP4RR#\R4h)NDYou cannot specify both input_ids and inputs_embeds at the same time5You have to specify either input_ids or inputs_embedsr&)rqr)rDrrs&&&rG_get_input_shape LayoutLMv2Model._get_input_shapesT  ]%>cd d  ">># #  & %%', ,TU UrIc^<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[S[,R S[S[,/ #) rPrNrUr^rrrBrrr)r? LongTensor FloatTensorrrtupler )rrgs"rGrLayoutLMv2Model.__annotate__ss s ##d*s %s   4' s ))D0 s ((4/ s &&-s ((4/s +,s  + +s rIc VPW4p Ve VPM VPp \V 4p VPP^,VPP^,,V ^&\ P !V 4p \VPW44p V ^;;,V ^,, uu&\ P !V 4p VPVPPW*V 4p \ P!W-.^R7pVf\ P!WR7p\ P!WPV R7p\ P!WO.^R7pVf(\ P!V \ PV R7pVf<V ^,pVPPRRV13,pVPV 4p\ P !^V ^,\ PV R7P#V ^,^4p\ P!VV.^R7pVfB\ P!\%\V 4^.,4\ PV R7pVP'VVVVVR7pVP)VV VR7p\ P!VV.^R7pVP+^4P+^4pVP-VPR7pR V, \ P.!VP4P0,pVP2!VV3R VR V/VBpVP4pVP7V4p\9VVR 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. image (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` or `detectron.structures.ImageList` whose `tensors` is of shape `(batch_size, num_channels, height, width)`): Batch of document images. Examples: ```python >>> from transformers import AutoProcessor, LayoutLMv2Model, set_seed >>> from PIL import Image >>> import torch >>> from datasets import load_dataset >>> set_seed(0) >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased") >>> model = LayoutLMv2Model.from_pretrained("microsoft/layoutlmv2-base-uncased") >>> dataset = load_dataset("hf-internal-testing/fixtures_docvqa") >>> image = dataset["test"][0]["image"] >>> encoding = processor(image, return_tensors="pt") >>> outputs = model(**encoding) >>> last_hidden_states = outputs.last_hidden_state >>> last_hidden_states.shape torch.Size([1, 342, 768]) ``` NrL)rirhrK)rrUrrBrr^rUrB)rrArUrB)r7 pooler_output)rrirrErr?SizerrTonesrr{rrrBrAr@rrrrr$rrrrr7rr )rDrrUr^rrrBrrrri visual_shaperr final_bboxvisual_attention_maskfinal_attention_maskrvisual_position_idsfinal_position_idstext_layout_emb visual_emb final_embextended_attention_maskencoder_outputssequence_outputrs&&&&&&&&, rGrLayoutLMv2Model.forwardsb++IE %.%:!! @T@TK( ++>>qADKKDhDhijDkk Qzz,/ 400JK A,q/)jj- ,,T[[-Q-QSWalm YY2:  !"ZZ CN % r)NN)NNNNNNN)r^r_r`rarri_can_record_outputsr(rrrrrrrrrrrcrdrerfs@@rGrNrN&sl*O\Kbc(/02 #JV s s  s s rIrNax LayoutLMv2 Model with a sequence classification head on top (a linear layer on top of the concatenation of the final hidden state of the [CLS] token, average-pooled initial visual embeddings and average-pooled final visual embeddings, 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 #)#LayoutLMv2ForSequenceClassificationic@<\SV`V4VPVn\V4Vn\ P !VP4Vn\ P!VP^,VP4Vn VP4R#)rRN r'r( num_labelsrNr<rr;r<r=rxr+ classifierrrCs&&rGr(,LayoutLMv2ForSequenceClassification.__init__sn   ++)&1zz&"<"<= ))F$6$6$:F\54pV!VP7RVP*4VP7R44pM.VP P(R8Xd\94pV!VV4p\;VVVP<VP>R7#)a input_ids (`torch.LongTensor` of shape `batch_size, sequence_length`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) 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. image (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` or `detectron.structures.ImageList` whose `tensors` is of shape `(batch_size, num_channels, height, width)`): Batch of document images. token_type_ids (`torch.LongTensor` of shape `batch_size, sequence_length`, *optional*): Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *sentence A* token, - 1 corresponds to a *sentence B* token. [What are token type IDs?](../glossary#token-type-ids) position_ids (`torch.LongTensor` of shape `batch_size, sequence_length`, *optional*): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. [What are position IDs?](../glossary#position-ids) 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). Example: ```python >>> from transformers import AutoProcessor, LayoutLMv2ForSequenceClassification, set_seed >>> from PIL import Image >>> import torch >>> from datasets import load_dataset >>> set_seed(0) >>> dataset = load_dataset("aharley/rvl_cdip", split="train", streaming=True) >>> data = next(iter(dataset)) >>> image = data["image"].convert("RGB") >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased") >>> model = LayoutLMv2ForSequenceClassification.from_pretrained( ... "microsoft/layoutlmv2-base-uncased", num_labels=dataset.info.features["label"].num_classes ... ) >>> encoding = processor(image, return_tensors="pt") >>> sequence_label = torch.tensor([data["label"]]) >>> outputs = model(**encoding, labels=sequence_label) >>> loss, logits = outputs.loss, outputs.logits >>> predicted_idx = logits.argmax(dim=-1).item() >>> predicted_answer = dataset.info.features["label"].names[4] >>> predicted_idx, predicted_answer # results are not good without further fine-tuning (7, 'advertisement') ``` NrrrhrrrUr^rrrBrrKrL regressionsingle_label_classificationmulti_label_classificationlosslogitsrrr&r])rKrOrK) rq%warn_if_padding_and_no_attention_maskrrirrErr?rr<rr@rrrr7r?rTr=r  problem_typer rrsrsqueezerrrr rr)rDrrUr^rrrBrrrrrirrrrinitial_image_embeddingsoutputsrrfinal_image_embeddingscls_final_outputpooled_initial_image_embeddingspooled_final_image_embeddingsrrloss_fcts&&&&&&&&&, rGr+LayoutLMv2ForSequenceClassification.forward&s^  ]%>cd d  "  6 6y Q#..*K  &',,.s3KTU U%.%:!! @T@TK( ++>>qADKKDhDhijDkk Qzz,/ ;' A,q/)jj- oo77 KK 0 0$  $ll1l1oUZZX^_ff NA $(??#G#G,$H$  /3oo /  /  /  / * / * / & / ( /  /   #..*K',,.s3K ^  % %a*n 5  % %an 50 +73+C*G*GA*G*N'(>(C(C(C(J%)) >@] ^de ,,71  {{''/??a'/;DKK,__q(fllejj.HFLL\a\e\eLe/LDKK,/KDKK,{{''<7"9??a'#FNN$4fnn6FGD#FF3D))-JJ+- B @&++b/R))-II,./'!//))   rIr r=r<r NNNNNNNN r^r_r`rar(rrrrrcrdrerfs@@rGrrs9:k k k k rIra LayoutLMv2 Model with a token classification head on top (a linear layer on top of the text part of the hidden states) e.g. for sequence labeling (information extraction) tasks such as [FUNSD](https://guillaumejaume.github.io/FUNSD/), [SROIE](https://rrc.cvc.uab.es/?ch=13), [CORD](https://github.com/clovaai/cord) and [Kleister-NDA](https://github.com/applicaai/kleister-nda). cbaa]tRtRtoV3RltRt]]RV3RlRll44tRt Vt V;t #) LayoutLMv2ForTokenClassificationic2<\SV`V4VPVn\V4Vn\ P !VP4Vn\ P!VPVP4Vn VP4R#rr rCs&&rGr()LayoutLMv2ForTokenClassification.__init__si   ++)&1zz&"<"<= ))F$6$68I8IJ rIcBVPPP#rrrs&rGr5LayoutLMv2ForTokenClassification.get_input_embeddingsrrIc<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[PR,R S[S[,R S[S[,/ #r)r?rrrrrr )rrgs"rGr-LayoutLMv2ForTokenClassification.__annotate__su u ##d*u %u   4' u ))D0 u ((4/ u &&-u ((4/u   4'u +,u  & &u rIc  VP!R RVRVRVRVRVRVRV/V Bp VeVP4p MVP4RR p V ^,p V PR RV 13,p VPV 4p VP V 4pRpVe=\ 4pV!VP R VP4VP R 44p\VVV PV PR 7#) a input_ids (`torch.LongTensor` of shape `batch_size, sequence_length`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) 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. image (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` or `detectron.structures.ImageList` whose `tensors` is of shape `(batch_size, num_channels, height, width)`): Batch of document images. token_type_ids (`torch.LongTensor` of shape `batch_size, sequence_length`, *optional*): Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *sentence A* token, - 1 corresponds to a *sentence B* token. [What are token type IDs?](../glossary#token-type-ids) position_ids (`torch.LongTensor` of shape `batch_size, sequence_length`, *optional*): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. [What are position IDs?](../glossary#position-ids) 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]`. Example: ```python >>> from transformers import AutoProcessor, LayoutLMv2ForTokenClassification, set_seed >>> from PIL import Image >>> from datasets import load_dataset >>> set_seed(0) >>> datasets = load_dataset("nielsr/funsd", split="test") >>> labels = datasets.features["ner_tags"].feature.names >>> id2label = {v: k for v, k in enumerate(labels)} >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr") >>> model = LayoutLMv2ForTokenClassification.from_pretrained( ... "microsoft/layoutlmv2-base-uncased", num_labels=len(labels) ... ) >>> data = datasets[0] >>> image = Image.open(data["image_path"]).convert("RGB") >>> words = data["words"] >>> boxes = data["bboxes"] # make sure to normalize your bounding boxes >>> word_labels = data["ner_tags"] >>> encoding = processor( ... image, ... words, ... boxes=boxes, ... word_labels=word_labels, ... padding="max_length", ... truncation=True, ... return_tensors="pt", ... ) >>> outputs = model(**encoding) >>> logits, loss = outputs.logits, outputs.loss >>> predicted_token_class_ids = logits.argmax(-1) >>> predicted_tokens_classes = [id2label[t.item()] for t in predicted_token_class_ids[0]] >>> predicted_tokens_classes[:5] # results are not good without further fine-tuning ['I-HEADER', 'I-HEADER', 'I-QUESTION', 'I-HEADER', 'I-QUESTION'] ``` rrUr^rrrBrNrKrr]r&) r<rr7r=r rrr r rr)rDrrUr^rrrBrrrr rrrrrr%s&&&&&&&&&, rGr(LayoutLMv2ForTokenClassification.forwards p/3oo /  /  /  / * / * / & / ( /  /   #..*K',,.s3K ^ !33A{ {NC,,71  ')HFKKDOO>> from transformers import AutoProcessor, LayoutLMv2ForQuestionAnswering, set_seed >>> import torch >>> from PIL import Image >>> from datasets import load_dataset >>> set_seed(0) >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv2-base-uncased") >>> model = LayoutLMv2ForQuestionAnswering.from_pretrained("microsoft/layoutlmv2-base-uncased") >>> dataset = load_dataset("hf-internal-testing/fixtures_docvqa") >>> image = dataset["test"][0]["image"] >>> question = "When is coffee break?" >>> encoding = processor(image, question, return_tensors="pt") >>> outputs = model(**encoding) >>> predicted_start_idx = outputs.start_logits.argmax(-1).item() >>> predicted_end_idx = outputs.end_logits.argmax(-1).item() >>> predicted_start_idx, predicted_end_idx (30, 191) >>> predicted_answer_tokens = encoding.input_ids.squeeze()[predicted_start_idx : predicted_end_idx + 1] >>> predicted_answer = processor.tokenizer.decode(predicted_answer_tokens) >>> predicted_answer # results are not good without further fine-tuning '44 a. m. to 12 : 25 p. m. 12 : 25 to 12 : 58 p. m. 12 : 58 to 4 : 00 p. m. 2 : 00 to 5 : 00 p. m. coffee break coffee will be served for men and women in the lobby adjacent to exhibit area. please move into exhibit area. ( exhibits open ) trrf general session ( part | ) presiding : lee a. waller trrf vice president " introductory remarks " lee a. waller, trrf vice presi - dent individual interviews with trrf public board members and sci - entific advisory council mem - bers conducted by trrf treasurer philip g. kuehn to get answers which the public refrigerated warehousing industry is looking for. plus questions from' ``` ```python >>> target_start_index = torch.tensor([7]) >>> target_end_index = torch.tensor([14]) >>> outputs = model(**encoding, start_positions=target_start_index, end_positions=target_end_index) >>> predicted_answer_span_start = outputs.start_logits.argmax(-1).item() >>> predicted_answer_span_end = outputs.end_logits.argmax(-1).item() >>> predicted_answer_span_start, predicted_answer_span_end (30, 191) ``` rrUr^rrrBrNrKrL) ignore_index)r start_logits end_logitsrrr]r&) r<rr7r7splitrrrGclamprr rr)rDrrUr^rrrBrr<r=rr rrrrrArB total_loss ignored_indexr% start_lossend_losss&&&&&&&&&&, rGr&LayoutLMv2ForQuestionAnswering.forwardysn/3oo /  /  /  / * / * / & / ( /  /   #..*K',,.s3K ^ !33A{ {NC1#)<<r<#: j#++B/::< ''+668   &=+D?'')*Q."1"9"9""==%%'(1, - 5 5b 9 (--a0M-33A}EO)//=AM']CH!,@J M:H$x/14J+%!!//))   rI)r<r r7)T) NNNNNNNNNr)rfs@@rGr5r5fs9 :E E E E rIr5)T r)r5rr+rrNr;)Erbrr?rtorch.nnrrrrrC activationsrmodeling_layersrmodeling_outputsr r r r r modeling_utilsrprocessing_utilsr pytorch_utilsrutilsrrrr utils.genericrrrutils.output_capturingrconfiguration_layoutlmv2rrQdetectron2.modelingrModule_load_from_state_dictrR batch_normrS get_loggerr^rrrirrrrrrrr;rsrFrrNrr+r5__all__r]rIrGr^s AA&!9.&6XX]]566LQ88??KpKpJ  22H   H %*+299*+ZP.biiP.f ")) 4 299  RYY  ryy "0"J)XT@ T@nAAA<09jryy9jx ryy f /f f R{ *C{ { |E '@E E PY %>Y Y x rI