+ jxRt^RIt^RIt^RIt^RIHt^RIHuHt^RIH t H t H t ^RI H t^RIHt^RIHt^RIHt^RIHtHtHtHt^RIHt^R IHt^R IHt^R I H!t!H"t"H#t#H$t$^R I%H&t&H't'^R I(H)t)^RI*H+t+]#PX!]-4t.!RR]P^4t0!RR]P^4t1!RR]P^4t2!RR]P^4t3!RR]P^4t4!RR]4t5!RR]P^4t6!RR]P^4t7!RR ]P^4t8]!!R!R"]44t9]!!R#R$]944t:!R%R&]P^4t;]!!R'R(7!R)R*]944t<]!!R+R,]944t=]!!R-R(7!R.R/]944t>.R0Ot?R#)1zPyTorch LayoutLMv3 model.N)BCEWithLogitsLossCrossEntropyLossMSELoss)initialization)ACT2FN)create_bidirectional_mask)GradientCheckpointingLayer)BaseModelOutputQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)PreTrainedModel)Unpack)apply_chunking_to_forward)auto_docstringcan_return_tuplelogging torch_int)TransformersKwargsmerge_with_config_defaults)capture_outputsLayoutLMv3Configc@aa]tRt^3toRtV3RltRRltRtVtV;t #)LayoutLMv3PatchEmbeddingszLayoutLMv3 image (patch) embeddings. This class also automatically interpolates the position embeddings for varying image sizes.cN<\SV`4\VP\P P 4'd VPMVPVP3p\VP\P P 4'd VPMVPVP3pV^,V^,,V^,V^,,3Vn\P!VPVPW3R7Vn R#)) kernel_sizestrideN)super__init__ isinstance input_size collectionsabcIterable patch_size patch_shapennConv2d num_channels hidden_sizeproj)selfconfig image_sizer& __class__s&& s/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/layoutlmv3/modeling_layoutlmv3.pyr "LayoutLMv3PatchEmbeddings.__init__7s &++[__-E-EFF   ##V%6%67 &++[__-E-EFF   ##V%6%67  'qMZ]:JqMZXY]S>STzz&"<"<= IIell63Q3Q&R&Y&YZa&botu!..#%<<  * *F,>,>DL\L\$  &(\\&2S2SU[UkUk%l"%'\\&2S2SU[UkUk%l"%'\\&2S2SU[UfUf%g"%'\\&2S2SU[UfUf%g"r3c2VPVR,4pVPVR,4pVPVR,4pVPVR,4pTP\P !TR,TR,, ^R44pTP \P !TR,TR,, ^R44p\P!TTTTTT.R R7p T # \dp\R4ThRp?ii;i) NNNz;The `bbox` coordinate values should be within 0-1000 range.Nidimryryr)ryryr\)ryryryryr8)rqrr IndexErrorrtricliprucat) r-bboxleft_position_embeddingsupper_position_embeddingsright_position_embeddingslower_position_embeddingsertruspatial_position_embeddingss && r1%calculate_spatial_position_embeddings>LayoutLMv3TextEmbeddings.calculate_spatial_position_embeddingsps c'+'A'A$w-'P $(,(B(B4=(Q %(,(B(B4=(Q %(,(B(B4=(Q %!% : :5::d7mVZ[bVcFcefhl;m n $ : :5::d7mVZ[bVcFcefhl;m n',ii()))%%   ' #+*% cZ[ab b csA C:: D DDc VPV4P4p\P!V^R7P V4V,pVP 4V,#)z Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols are ignored. This is modified from fairseq's `utils.make_positions`. rz)neintricumsumtype_aslong)r- input_idsrVmaskincremental_indicess&&& r1"create_position_ids_from_input_ids;LayoutLMv3TextEmbeddings.create_position_ids_from_input_idssP ||K(,,.$||Da8@@F$N"'')K77r3c .VP4RRpV^,p\P!VP^,W0P,^,\PVP R7pVP ^4PV4#)zq We are provided embeddings directly. We cannot infer which are padded so just generate sequential position ids. Ndtypedevicer8)r6rirjrVrr unsqueezerl)r- inputs_embeds input_shapesequence_lengthrms&& r1&create_position_ids_from_inputs_embeds?LayoutLMv3TextEmbeddings.create_position_ids_from_inputs_embedss~$((*3B/ %a.||   q /4D4D"Dq"HPUPZPZcpcwcw %%a(// < J": "66|D) &*&P&PQU&V#= ^^J/ \\*- r3) rcrgrtrVrnrmrbrur`rqrrNNNNN) rIrJrKrLrMr rrrrErNrOrPrQs@@r1rTrTVs)h*+48 =''r3rTcRaa]tRt^toV3RltRRltRV3RlRlltRtVtV;t #)LayoutLMv3SelfAttentioncB<\SV`4VPVP,^8wd9\ VR4'g'\ RVP RVP R24hVPVn\ VPVP, 4VnVPVP,Vn\P!VPVP4Vn \P!VPVP4Vn \P!VPVP4Vn \P!VP4VnVP"VnVP$VnR#)rembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ()N)rr r+num_attention_headshasattr ValueErrorrattention_head_size all_head_sizer(Linearquerykeyvaluereattention_probs_dropout_probrghas_relative_attention_biashas_spatial_attention_biasrvs&&r1r LayoutLMv3SelfAttention.__init__s;     : : :a ?PVXhHiHi#F$6$6#78 445Q8  $*#=#= #&v'9'9FS"S%**,CD --r3) rrrgrrrrrr) NNN) rIrJrKrLr rrErNrOrPrQs@@r1rrsL( 85.5.5.r3rcDaa]tRtRtoV3RltV3RlRltRtVtV;t#)LayoutLMv3SelfOutputi$c0<\SV`4\P!VPVP4Vn\P !VPVPR7Vn\P!VP4Vn R#rWN) rr r(rr+densercrdrerfrgrvs&&r1r LayoutLMv3SelfOutput.__init__%s` YYv1163E3EF f&8&8f>S>STzz&"<"<= r3ch<V^8dQhRS[PRS[PRS[P/#r}r input_tensorreturnriTensor)rrRs"r1r!LayoutLMv3SelfOutput.__annotate__+1U\\RWR^R^r3cxVPV4pVPV4pVPW,4pV#rGrrgrcr-rrs&&&r1rELayoutLMv3SelfOutput.forward+5 =1  ]3 }'CD r3rcrrg rIrJrKrLr rErNrOrPrQs@@r1rr$> r3rcHaa]tRtRtoV3RltRV3RlRlltRtVtV;t#)LayoutLMv3Attentioni3cd<\SV`4\V4Vn\ V4VnR#rG)rr rr-routputrvs&&r1r LayoutLMv3Attention.__init__4s& +F3 *62 r3c0<V^8dQhRS[S[,/#rr)rrRs"r1r LayoutLMv3Attention.__annotate__9s  +,  r3c `TpVP!VV3RVRV/VBwrxVPWv4pV#rr)r-r) r-rrrrrresidualattention_output_s &&&&&, r1rELayoutLMv3Attention.forward9sW!"ii    "     ;;'7Br3)rr-rrrQs@@r1rr3s3    r3rcNaa]tRtRtoV3RltRV3RlRlltRtRtVtV;t #)LayoutLMv3LayeriNc<\SV`4VPVn^Vn\ V4Vn\ V4Vn\V4Vn R#)r\N) rr chunk_size_feed_forward seq_len_dimr attentionLayoutLMv3Intermediate intermediateLayoutLMv3Outputrrvs&&r1r LayoutLMv3Layer.__init__OsI '-'E'E$,V426:&v. r3c0<V^8dQhRS[S[,/#rr)rrRs"r1rLayoutLMv3Layer.__annotate__Ws+,r3c VPVVVVR7p\VPVPVPV4pV#)r)rrfeed_forward_chunkrr) r-rroutput_attentionsrrrr layer_outputs &&&&&&, r1rELayoutLMv3Layer.forwardWsT >>  ! * 1  # #T%A%A4CSCSUe r3cJVPV4pVPW!4pV#rG)rr)r-rintermediate_outputrs&& r1r"LayoutLMv3Layer.feed_forward_chunkms)"//0@A{{#6I r3)rrrrr)NFNN) rIrJrKrLr rErrNrOrPrQs@@r1rrNs/,r3rc^aa]tRtRtoV3RltR RltRtRtR V3RlRlltRt Vt V;t #) LayoutLMv3Encoderisc<\SV`4Wn\P!\ VP 4Uu.uFp\V4NK up4VnRVn VPVn VPVn VP'dUVPVn VPVn \P!VPVPRR7VnVP'dVP"VnVP$Vn\P!VP$VPRR7Vn\P!VP$VPRR7VnR#R#uupi)F)biasN)rr r.r( ModuleListrangenum_hidden_layersrlayergradient_checkpointingrr rel_pos_bins max_rel_posrr rel_pos_biasmax_rel_2d_posrel_2d_pos_binsrel_pos_x_biasrel_pos_y_bias)r-r.rr0s&& r1r LayoutLMv3Encoder.__init__ts)  ]]U6KcKcEd#eEdOF$;Ed#ef &+#+1+M+M(*0*K*K'  + + + & 3 3D %11D  " $*;*;V=W=W^c dD   * * *"("7"7D #)#9#9D "$))D,@,@&B\B\ch"iD "$))D,@,@&B\B\ch"iD  +$fsFc^pV'dAV^,pWQ^8P4V,, p\P!V4pM,\P!V)\P!V44pV^,pWg8pV\P !VP 4V, 4\P !WG, 4, W7, ,P\P4,p \P!V \P!W^, 44p V\P!WV 4, pV#)r) rriabsmax zeros_likelogfloatrrmin full_likewhere) r-relative_position bidirectional num_buckets max_distanceretn max_exactis_small val_if_larges &&&&& r1relative_position_bucket*LayoutLMv3Encoder.relative_position_buckets  A K )//1K? ?C +,A ,,e.>.>?P.QRA 1$ =! IIaggi)+ ,txx 8P/Q QU`Ul m "UZZ. yyu|[\_/]^  u{{8 55 r3cVPR4VPR4, pVPVVPVPR7p\P !4;_uu_4VP PP4V,P^^^^4pRRR4VP4pV# +'giL";i)r}r1r2Nrr8) rr8rrrino_gradr weighttr:r)r-rm rel_pos_matrs&& r1_cal_1d_pos_emb!LayoutLMv3Encoder._cal_1d_pos_embs",,R0<3I3I"3MM // ))))0 ]]__''..0027;CCAq!QOG$$&_s ->C C cVR,pVR,pVPR4VPR4, pVPR4VPR4, pVPVVPVPR7pVPVVPVPR7p\P !4;_uu_4VP PP4V,P^^^^4pVPPP4V,P^^^^4pRRR4VP4pVP4pWg,pV# +'giL:;i)ryr;Nr|r~rr8) rr8r"r!rir<r#r=r>r:r$r) r-rposition_coord_xposition_coord_yrel_pos_x_2d_matrel_pos_y_2d_mat rel_pos_x rel_pos_yrs && r1_cal_2d_pos_emb!LayoutLMv3Encoder._cal_2d_pos_embs\==+55b9S>STzz&"<"<= r3ch<V^8dQhRS[PRS[PRS[P/#rr)rrRs"r1rLayoutLMv3Output.__annotate__rr3cxVPV4pVPV4pVPW,4pV#rGrrs&&&r1rELayoutLMv3Output.forwardrr3rrrQs@@r1r r rr3r czaa]tRtRtoRtRtR]R]/t] P!4V3Rl4t V3Rlt Rt VtV;t#) LayoutLMv3PreTrainedModeli layoutlmv3r attentionsc <\SV`V4\V\4'dVPP 'dA\ P!VP4\ P!VP4\VR4'dJ\ P!VPVPVPVP3R74R#R#\V\4'db\ P!VP \"P$!VP P&R,4P)R44R#R#)zInitialize the weights visual_bboxr/Nr8r[)r _init_weightsr!LayoutLMv3Modelr. visual_embedinitzeros_ cls_token pos_embedrcopy_rkcreate_visual_bboxr6rTrmrirjr;rl)r-moduler0s&&r1rm'LayoutLMv3PreTrainedModel._init_weights s f% fo . .{{''' F,,- F,,-v}-- 6--v/H/HU[U`U`bhbmbmTn/H/op.  8 9 9 JJv**ELL9L9L9R9RSU9V,W,^,^_f,g h:r3c&<V^8dQh/S[;R&#)r}r.r)rrRs"r1r&LayoutLMv3PreTrainedModel.__annotate__s r3)imagetext)rIrJrKrLbase_model_prefixinput_modalitiesrr_can_record_outputsrir<rm__annotate_func__rNrOrPrQs@@r1rgrgsD%(*O\Kbc ]]_ i ir3rgcaa]tRtRtoV3RltRtRtR RltRtRt ] ] ] R V3RlR ll444t R tVtV;t#) rnic<\SV`V4WnVP'd\ V4VnVP 'Ed\V4Vn\VPVP, 4Vn \P!\P !^^VP"44Vn\P!\P !^VPVP,^,VP"44Vn\P(!RR7Vn\P,!VP"VP.R7Vn\P(!VP04VnVPP4'gVPP6'dD\P8!VP;VPVP3R7RR7Vn\P,!VP"RR7Vn\AV4Vn!VPE4R#) r\g)prWrlFrYgư>N)#rr r. text_embedrTrBror patch_embedrr"r&r6r( Parameterrirr+rrrsrepos_droprcrdrfrgrrrhrurknormrencoder post_initrvs&&r1r LayoutLMv3Model.__init__s      6v>DO     9@D F--0A0AABDI\\%++aFVPV4pVP4wr4pVPPVRR4p\P !Wb3^R7pVP eW P ,pVPV4pVPV4pV#)r\rzr8) rr6rrrlrirrsrr)r-r@rBrseq_lenr cls_tokenss&& r1 forward_imageLayoutLMv3Model.forward_image[s%%l3 ",!2 Q^^**:r2> YY 7Q?  >> %#nn4J]]:. YYz* r3c^<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[S[,R S[S[,/ #) r}rNrrrrmrr@rr)ri LongTensor FloatTensorrrtupler )rrRs"r1rLayoutLMv3Model.__annotate__osm m ##d*m %m ))D0 m ((4/ m &&- m ((4/m ''$.m +,m   m r3c  Ve"VP4p V wrVPp MOVe%VP4RRp V wrVPp M'Ve\V4p VPp M \R4hVfVeVf\P !V X 3V R7pVf(\P !X \PV R7pVfB\P !\\X 4^.,4\PV R7pVPVVVVVR7p R;rR;ppVEei\VP^,VPP, 4\VP^,VPP, 4ppVPV4p\P !V VP^,3\PV R7pVe\P !VV.^R7pMTpVPP"'gVPP$'EdVPP$'dDVP'V \PV R7pVe\P !VV.^R7pMTp\P(!^VP^,\PV R7P+V ^4pVfVe\\P(!^X ^,V R7P-^4pVP/V 4p\P !VV.^R7pMTpVfVe\P !X V.^R7p MTp VP1V 4p VP3V 4p MVPP"'gVPP$'dvVPP$'dTpVPP"'d<VPP4RRX ^,13,pVP7V4pTp\9VPX VR 7pVP:!V 3R VR VR VR VRV/VBpVP<p\?VR7#)a$ input_ids (`torch.LongTensor` of shape `(batch_size, token_sequence_length)`): Indices of input sequence tokens in the vocabulary. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. 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, token_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. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. token_type_ids (`torch.LongTensor` of shape `(batch_size, token_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. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. [What are token type IDs?](../glossary#token-type-ids) position_ids (`torch.LongTensor` of shape `(batch_size, token_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]`. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. [What are position IDs?](../glossary#position-ids) inputs_embeds (`torch.FloatTensor` of shape `(batch_size, token_sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert *input_ids* indices into associated vectors than the model's internal embedding lookup matrix. Examples: ```python >>> from transformers import AutoProcessor, AutoModel >>> from datasets import load_dataset >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = AutoModel.from_pretrained("microsoft/layoutlmv3-base") >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train") >>> example = dataset[0] >>> image = example["image"] >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> encoding = processor(image, words, boxes=boxes, return_tensors="pt") >>> outputs = model(**encoding) >>> last_hidden_states = outputs.last_hidden_state ```NzEYou have to specify either input_ids or inputs_embeds or pixel_values)rr)rrrmrrrz)rrry)r.rrrrmrrCrDrNr8) r6rlenrrionesrrrlistrBrr;r.r&rrrrrrjrrrlrcrgrm expand_asrrrOr )r-rrrrrmrr@rrr seq_lengthrembedding_output final_bboxfinal_position_idsrCrDvisual_embeddingsvisual_attention_maskrkvisual_position_idsencoder_outputssequence_outputs&&&&&&&&, r1rELayoutLMv3Model.forwardls Z  #..*K%0 "J%%F  &',,.s3K%0 "J"))F  %\*J!((Fde e  M$=%!&j*-Ev!V%!&[ SY!Z|{{5k):aS)@#A\bc##)-+ /  +/. %)) {  #,,,Q/$++2H2HHI,,,Q/$++2H2HHI&L!% 2 2< @ $)JJ.44Q78 SY% !)!&N&#Q#[#[\]#^L#/#6#6{#CL).LBU3V\])^&)<&$ (A#(99.>@Q-RXY#Z #4 #~~.>? #||,<=  [[ 4 4 4 8^8^8^{{555! {{666#;;A?OQ?O/0>. " m m  m m r3rnc@aa]tRtRtoRtRV3RlltRtRtVtV;t #)LayoutLMv3ClassificationHeadizT Head for sentence-level classification tasks. Reference: RobertaClassificationHead c<\SV`4W nV'd9\P!VP ^,VP 4VnM0\P!VP VP 4VnVPe VPM VPp\P!V4Vn \P!VP VP4Vn R#)rN) rr pool_featurer(rr+rclassifier_dropoutrfrerg num_labelsout_proj)r-r.rrr0s&&& r1r %LayoutLMv3ClassificationHead.__init__$s ( 6#5#5#96;M;MNDJ6#5#5v7I7IJDJ)/)B)B)NF % %TZTnTn zz"45  &"4"4f6G6GH r3cVPV4pVPV4p\P!V4pVPV4pVP V4pV#rG)rgrritanhr)r-xs&&r1rE$LayoutLMv3ClassificationHead.forward1sI LLO JJqM JJqM LLO MM! r3)rrgrr)FrHrQs@@r1rrs Ir3ra LayoutLMv3 Model with a token classification head on top (a linear layer on top of the final 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). ) custom_introchaa]tRtRtoV3RltRtRt]]RV3RlRll44t Rt Vt V;t #) LayoutLMv3ForTokenClassificationi:cz<\SV`V4VPVn\V4Vn\ P !VP4VnVP^ 8d2\ P!VPVP4Vn M\VRR7Vn VP4R#) FrN)rr rrnrhr(rerfrgrr+ classifierrrrvs&&r1r )LayoutLMv3ForTokenClassification.__init__Cs   ++)&1zz&"<"<=   r ! ii(:(:F>> from transformers import AutoProcessor, AutoModelForTokenClassification >>> from datasets import load_dataset >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = AutoModelForTokenClassification.from_pretrained("microsoft/layoutlmv3-base", num_labels=7) >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train") >>> example = dataset[0] >>> image = example["image"] >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> word_labels = example["ner_tags"] >>> encoding = processor(image, words, boxes=boxes, word_labels=word_labels, return_tensors="pt") >>> outputs = model(**encoding) >>> loss = outputs.loss >>> logits = outputs.logits ```rrrrmrr@Nrylosslogitsrrir8) rhr6rgrrr9rr rri)r-rrrrrmrrr@routputsrrrrrloss_fcts&&&&&&&&&, r1rE(LayoutLMv3ForTokenClassification.forwardVsZ//     *  *  &  (  &      #..*K',,.s3K ^ !!*Q  ^4,,71  ')HFKKDOO 64J J J J r3rchaa]tRtRtoV3RltRtRt]]RV3RlRll44t Rt Vt V;t #) LayoutLMv3ForQuestionAnsweringic<\SV`V4VPVn\V4Vn\ VRR7VnVP4R#FrN)rr rrnrhr qa_outputsrrvs&&r1r 'LayoutLMv3ForQuestionAnswering.__init__sA   ++)&16vER r3c6VPP4#rGrrs&r1r3LayoutLMv3ForQuestionAnswering.get_input_embeddingsrr3c<VPPV4R#rGrrs&&r1r3LayoutLMv3ForQuestionAnswering.set_input_embeddingsrr3c<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[PR,R S[PR,R S[S[,R S[S[,/ #) r}rNrrrmrstart_positions end_positionsrr@rr)rirrrrrr )rrRs"r1r+LayoutLMv3ForQuestionAnswering.__annotate__sW W ##d*W ))D0W ((4/ W &&- W ((4/ W ))D0W ''$.W %W &&-W +,W  - -W r3c  VP!V3RVRVRVRVRVRV /V Bp V ^,p VPV 4p V P^R R7wrVPR 4P 4pVPR 4P 4pRpVeVe\ VP 44^8dVPR 4p\ VP 44^8dVPR 4pVP ^4pVP^V4pVP^V4p\VR 7pV!W4pV!W4pVV,^, p\VVVV PV PR 7#) aG bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. Examples: ```python >>> from transformers import AutoProcessor, AutoModelForQuestionAnswering >>> from datasets import load_dataset >>> import torch >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = AutoModelForQuestionAnswering.from_pretrained("microsoft/layoutlmv3-base") >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train") >>> example = dataset[0] >>> image = example["image"] >>> question = "what's his name?" >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> encoding = processor(image, question, words, boxes=boxes, return_tensors="pt") >>> start_positions = torch.tensor([1]) >>> end_positions = torch.tensor([3]) >>> outputs = model(**encoding, start_positions=start_positions, end_positions=end_positions) >>> loss = outputs.loss >>> start_scores = outputs.start_logits >>> end_scores = outputs.end_logits ```rrrmrrr@rzN) ignore_index)r start_logits end_logitsrrir8) rhrsplitsqueezerrr6clamprr rri)r-rrrrmrrrrr@rrrrrr total_loss ignored_indexr start_lossend_losss&&&&&&&&&&, r1rE&LayoutLMv3ForQuestionAnswering.forwards`$(??  $ ) $ * $ & $ ( $  $ & $  $ "!*1#)<<r<#: #++B/::< ''+668   &=+D?'')*Q."1"9"9""==%%'(1, - 5 5b 9 (--a0M-33A}EO)//=AM']CH!,@J :H$x/14J+%!!//))   r3)rhrr) NNNNNNNNNrrQs@@r1rrs>64W W W W r3ra  LayoutLMv3 Model with a sequence classification head on top (a linear layer on top of the final hidden state of the [CLS] token) e.g. for document image classification tasks such as the [RVL-CDIP](https://www.cs.cmu.edu/~aharley/rvl-cdip/) dataset. chaa]tRtRtoV3RltRtRt]]RV3RlRll44t Rt Vt V;t #) #LayoutLMv3ForSequenceClassificationic<\SV`V4VPVnWn\ V4Vn\ VRR7VnVP4R#r) rr rr.rnrhrrrrvs&&r1r ,LayoutLMv3ForSequenceClassification.__init__sF   ++ )&16vER r3c6VPP4#rGrrs&r1r8LayoutLMv3ForSequenceClassification.get_input_embeddings#rr3c<VPPV4R#rGrrs&&r1r8LayoutLMv3ForSequenceClassification.set_input_embeddings&rr3c<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}rNrrrmrrrr@rr)rirrrrrr )rrRs"r1r0LayoutLMv3ForSequenceClassification.__annotate__+sT T ##d*T ))D0T ((4/ T &&- T ((4/ T   4'T %T &&-T +,T  ) )T r3c  VP!V3RVRVRVRVRVRV/V Bp V ^,R ,p VPV 4p Rp VEeVPPfVP^8XdRVPnMrVP^8dQVP \ P8Xg VP \ P8XdR VPnMR VPnVPPR8XdK\4pVP^8Xd'V!V P4VP44p MV!W4p MVPPR 8Xd>\4pV!V PR VP4VPR 44p M-VPPR 8Xd\4pV!W4p \V V V PV 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. Examples: ```python >>> from transformers import AutoProcessor, AutoModelForSequenceClassification >>> from datasets import load_dataset >>> import torch >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = AutoModelForSequenceClassification.from_pretrained("microsoft/layoutlmv3-base") >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train") >>> example = dataset[0] >>> image = example["image"] >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> encoding = processor(image, words, boxes=boxes, return_tensors="pt") >>> sequence_label = torch.tensor([1]) >>> outputs = model(**encoding, labels=sequence_label) >>> loss = outputs.loss >>> logits = outputs.logits ```rrrmrrr@N regressionsingle_label_classificationmulti_label_classificationr)ryrryr8)rhrr. problem_typerrrirrrrrr9rr rri)r-rrrrmrrrr@rrrrrrs&&&&&&&&&, r1rE+LayoutLMv3ForSequenceClassification.forward)sX$(??  $ ) $ * $ & $ ( $  $ & $  $ "!*W-1  {{''/??a'/;DKK,__q(fllejj.HFLL\a\e\eLe/LDKK,/KDKK,{{''<7"9??a'#FNN$4fnn6FGD#F3D))-JJ+- B @&++b/R))-II,./'!//))   r3)rr.rhrrrrQs@@r1rrs>64T T T T r3r)rrrrnrg)@rMr#rritorch.nnr(torch.nn.functional functionalr<rrrrrp activationsr masking_utilsrmodeling_layersrmodeling_outputsr r r r modeling_utilsr processing_utilsr pytorch_utilsrutilsrrrr utils.genericrrutils.output_capturingrconfiguration_layoutlmv3r get_loggerrIloggerModulerrTrrrrrrr rgrnrrrr__all__rzr3r1r$s  AA&!69 .&6 L56   H %  FqryyqhV.biiV.t 299  ")) 6"0"Jn@ n@d RYY  ryy iii(C /C C L2996` '@` ` Fi %>i i Xf *Cf f R r3