fThJSrSSKrSSKrSSKJrJrJr SSKrSSKJ r SSK J s J r SSK rSSKJrJrJr SSKJr SSKJrJrJrJr SSKJr SS KJr SS KJrJrJr S S K J!r! \RD"\#5r$"S S\ RJ5r&"SS\ RJ5r'\"SS\55r("SS\ RJ5r)"SS\ RJ5r*"SS\ RJ5r+"SS\ RJ5r,"SS\ RJ5r-"SS\ RJ5r."SS \ RJ5r/\"S!S"\(55r0"S#S$\ RJ5r1\"S%S&9"S'S(\(55r2\"S)S*\(55r3\"S+S&9"S,S-\(55r4/S.Qr5g)/zPyTorch LayoutLMv3 model.N)OptionalTupleUnion)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)BaseModelOutputQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)PreTrainedModel)apply_chunking_to_forward)auto_docstringlogging torch_int)LayoutLMv3Configc6^\rSrSrSrU4SjrSSjrSrU=r$)LayoutLMv3PatchEmbeddings/zLayoutLMv3 image (patch) embeddings. This class also automatically interpolates the position embeddings for varying image sizes.c>[TU]5 [UR[R R 5(a UROURUR4n[UR[R R 5(a UROURUR4nUSUS-USUS-4Ul[R"URURX3S9Ul g)Nrr) kernel_sizestride)super__init__ isinstance input_size collectionsabcIterable patch_size patch_shapennConv2d num_channels hidden_sizeproj)selfconfig image_sizer# __class__s j/var/www/auris/envauris/lib/python3.13/site-packages/transformers/models/layoutlmv3/modeling_layoutlmv3.pyr"LayoutLMv3PatchEmbeddings.__init__3s &++[__-E-EFF   ##V%6%67 &++[__-E-EFF   ##V%6%67  'qMZ]:JqMZXY][TU]5 [R"URUR UR S9Ul[R"URUR 5Ul [R"UR URS9Ul [R"UR5UlURS[ R""UR$5R'S5SS9 UR Ul[R"UR$UR UR(S9Ul[R"UR,UR.5Ul[R"UR,UR.5Ul[R"UR,UR45Ul[R"UR,UR45Ulg)N) padding_idxeps position_ids)rr2F) persistent)rrr% Embedding vocab_sizer( pad_token_idword_embeddingstype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutregister_buffertorcharangemax_position_embeddingsexpandrSposition_embeddingsmax_2d_position_embeddingscoordinate_sizex_position_embeddingsy_position_embeddings shape_sizeh_position_embeddingsw_position_embeddingsr*r+r-s r.r!LayoutLMv3TextEmbeddings.__init__Ws} !||F,=,=v?Q?Q_e_r_rs%'\\&2H2H&J\J\%]"f&8&8f>S>STzz&"<"<=   ELL)G)GHOOPWXej  "..#%<<  * *F,>,>DL\L\$  &(\\&2S2SU[UkUk%l"%'\\&2S2SU[UkUk%l"%'\\&2S2SU[UfUf%g"%'\\&2S2SU[UfUf%g"r0c JURUSS2SS2S45nURUSS2SS2S45nURUSS2SS2S45nURUSS2SS2S45nUR[R "USS2SS2S4USS2SS2S4- SS55nUR [R "USS2SS2S4USS2SS2S4- SS55n[R"UUUUUU/SS9n U $![an[S5UeSnAff=f) Nrrr3r z;The `bbox` coordinate values should be within 0-1000 range.ir2dim)rkrl IndexErrorrnrdcliprocat) r*bboxleft_position_embeddingsupper_position_embeddingsright_position_embeddingslower_position_embeddingsernrospatial_position_embeddingss r.%calculate_spatial_position_embeddings>LayoutLMv3TextEmbeddings.calculate_spatial_position_embeddingsnsN c'+'A'A$q!Qw-'P $(,(B(B41a=(Q %(,(B(B41a=(Q %(,(B(B41a=(Q %!% : :5::d1aQR7mVZ[\^_ab[bVcFcefhl;m n $ : :5::d1aQR7mVZ[\^_ab[bVcFcefhl;m n',ii()))%%   ' #+*% cZ[ab b csA,D D" DD"cURU5R5n[R"USS9R U5U-nUR 5U-$)z Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols are ignored. This is modified from fairseq's `utils.make_positions`. rrs)neintrdcumsumtype_aslong)r* input_idsrSmaskincremental_indicess r."create_position_ids_from_input_ids;LayoutLMv3TextEmbeddings.create_position_ids_from_input_idssP ||K(,,.$||Da8@@F$N"'')K77r0cUR5SSnUSn[R"URS-X0R-S-[RUR S9nUR S5RU5$)zq We are provided embeddings directly. We cannot infer which are padded so just generate sequential position ids. Nr2rdtypedevicer)r5rdrerSrr unsqueezerg)r* inputs_embeds input_shapesequence_lengthrVs r.&create_position_ids_from_inputs_embeds?LayoutLMv3TextEmbeddings.create_position_ids_from_inputs_embedss~$((*3B/ %a.||   q /4D4D"Dq"HPUPZPZcpcwcw %%a(// < J": "66|D) &*&P&PQU&V#= ^^J/ \\*- r0) r^rbrnrSrhr]ror[rkrl)NNNNN) rGrHrIrJrKrrrrrCrLrMrNs@r.rPrPRs8h.+48 =  ''r0rPc"\rSrSr\rSrSrSrg)LayoutLMv3PreTrainedModel layoutlmv3c,[U[R[R45(akURR R SURRS9 URb%URR R5 gg[U[R5(axURR R SURRS9 URb2URR URR5 gg[U[R5(aJURR R5 URR RS5 g[U[5(aeURR (aIUR"R R5 UR$R R5 ggg)zInitialize the weights)meanstdNg?)rr%Linearr&weightdatanormal_r+initializer_rangebiaszero_rXrSr^fill_LayoutLMv3Model visual_embed cls_token pos_embed)r*modules r. _init_weights'LayoutLMv3PreTrainedModel._init_weightssg fryy"))4 5 5 MM   & &CT[[5R5R & S{{&   &&('  - - MM   & &CT[[5R5R & S!!- ""6#5#56<<>.  - - KK   " " $ MM   $ $S )  0 0{{''  %%++-  %%++-(1r0N) rGrHrIrJr config_classbase_model_prefixrrLrr0r.rrs#L$.r0rcL^\rSrSrU4SjrSrSSjrSSjrSrU=r $) LayoutLMv3SelfAttentionc$>[TU]5 URUR-S:wa7[ US5(d&[ SURSURS35eURUl[ URUR- 5UlURUR-Ul[R"URUR5Ul [R"URUR5Ul [R"URUR5Ul [R"UR5UlUR"UlUR$Ulg)Nrembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ())rrr(num_attention_headshasattr ValueErrorrattention_head_size all_head_sizer%rquerykeyvaluer`attention_probs_dropout_probrbhas_relative_attention_biashas_spatial_attention_biasrps r.r LayoutLMv3SelfAttention.__init__s;     : : :a ?PVXhHiHi#F$6$6#78 445Q8  $*#=#= #&v'9'9FS"S%**,CD 6G=2O\M]r0) rrrbrrrrrr) NNFNN) rGrHrIrJrrrrCrLrMrNs@r.rrs.L(% 811r0rcz^\rSrSrU4SjrS\R S\R S\R 4SjrSrU=r $)LayoutLMv3SelfOutputi=c(>[TU]5 [R"URUR5Ul[R "URURS9Ul[R"UR5Ul gNrT) rrr%rr(denser^r_r`rarbrps r.rLayoutLMv3SelfOutput.__init__>s` YYv1163E3EF f&8&8f>S>STzz&"<"<= r0r input_tensorreturncpURU5nURU5nURX-5nU$rErrbr^r*rrs r.rCLayoutLMv3SelfOutput.forwardD5 =1  ]3 }'CD r0r^rrb rGrHrIrJrrdTensorrCrLrMrNs@r.rr=6> U\\RWR^R^r0rc<^\rSrSrU4SjrSSjrSrU=r$)LayoutLMv3AttentioniLcb>[TU]5 [U5Ul[ U5UlgrE)rrrr*routputrps r.rLayoutLMv3Attention.__init__Ms& +F3 *62 r0c jURUUUUUUS9nURUSU5nU4USS-n U $)Nrrrr)r*r) r*rrrrrr self_outputsattention_outputrs r.rCLayoutLMv3Attention.forwardRsYyy    ! !  ;;|A F#% QR(88r0)rr*r)rGrHrIrJrrCrLrMrNs@r.rrLs#3r0rcB^\rSrSrU4SjrSSjrSrSrU=r$)LayoutLMv3Layeriic>[TU]5 URUlSUl[ U5Ul[ U5Ul[U5Ul gNr) rrchunk_size_feed_forward seq_len_dimr attentionLayoutLMv3Intermediate intermediateLayoutLMv3Outputrrps r.rLayoutLMv3Layer.__init__jsI '-'E'E$,V426:&v. r0c URUUUUUUS9nUSnUSSn [URURURU5n U 4U -n U $)N)rrrrr)rrfeed_forward_chunkr r ) r*rrrrrrself_attention_outputsrr layer_outputs r.rCLayoutLMv3Layer.forwardrs|"&   /! "0" 2!4(,0  # #T%A%A4CSCSUe  /G+r0cJURU5nURX!5nU$rE)rr)r*rintermediate_outputrs r.r"LayoutLMv3Layer.feed_forward_chunks)"//0@A{{#6I r0)rr rrr r) rGrHrIrJrrCrrLrMrNs@r.r r is(/8r0r cZ^\rSrSrU4SjrSSjrSrSrS SjrSr U=r $) LayoutLMv3Encoderic>[TU]5 Xl[R"[ UR 5Vs/sHn[U5PM sn5UlSUl URUl URUl UR(aQURUl URUl [R"URURSS9UlUR(aUR"UlUR$Ul[R"UR$URSS9Ul[R"UR$URSS9Ulggs snf)NF)r)rrr+r% ModuleListrangenum_hidden_layersr layergradient_checkpointingrr rel_pos_bins max_rel_posrr rel_pos_biasmax_rel_2d_posrel_2d_pos_binsrel_pos_x_biasrel_pos_y_bias)r*r+_r-s r.rLayoutLMv3Encoder.__init__s#  ]]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 "$))D,@,@&B\B\ch"iD  +$fsE<cDSnU(a4US-nXQS:R5U-- n[R"U5nO,[R"U*[R"U55nUS-nXg:nU[R "UR 5U- 5[R "XG- 5- X7- -R[R5-n [R"U [R"XS- 55n U[R"XU 5- nU$)Nrr3r) rrdabsmax zeros_likelogfloatrrmin full_likewhere) r*relative_position bidirectional num_buckets max_distanceretn max_exactis_small val_if_larges r.relative_position_bucket*LayoutLMv3Encoder.relative_position_buckets  A K )//1K? ?C +,A ,,e.>.>?P.QRA 1$ =! IIaggi)+ ,txx 8P/Q QU`Ul m "UZZ. yyu|[\_/]^  u{{8 55 r0cURS5URS5- nURUURURS9n[R "5 UR RR5URSSSS5nSSS5 UR5nU$!,(df  N =f)Nrr2r7r8rr rr3) rr>r#r$rdno_gradr%rtr8r)r*rV rel_pos_matrs r._cal_1d_pos_emb!LayoutLMv3Encoder._cal_1d_pos_embs",,R0<3I3I"3MM // ))))0 ]]_''..0027;CCAq!QOG$$&_s :B22 CcUSS2SS2S4nUSS2SS2S4nURS5URS5- nURS5URS5- nURUURURS9nURUURURS9n[R "5 UR RR5URSSSS5nURRR5URSSSS5nSSS5 UR5nUR5nXg-nU$!,(df  N4=f)Nrr rr2rArr3) rr>r'r&rdrBr(rrCr8r)r) r*rxposition_coord_xposition_coord_yrel_pos_x_2d_matrel_pos_y_2d_mat rel_pos_x rel_pos_yrs r._cal_2d_pos_emb!LayoutLMv3Encoder._cal_2d_pos_embsf1a=1a=+55b9A3E E-c XU(aSOSn U(aSOSn UR(aURU5OSn UR(aURU5OSn[ UR 5HunnU(aX4-n UbXOOSnUR (a3UR(a"URURUUUUU U5nO U"UUUUU US9nUSnU(dMxU US4-n M U(aX4-n U(d[SUU U 455$[UU U S9$)Nrrrrc30# UH nUcMUv M g7frEr).0vs r. ,LayoutLMv3Encoder.forward..!s"A  s last_hidden_stater attentions) rrErrN enumerater!r"training_gradient_checkpointing_func__call__tupler )r*rrxrrroutput_hidden_states return_dictrVrArBall_hidden_statesall_self_attentionsrri layer_modulelayer_head_mask layer_outputss r.rCLayoutLMv3Encoder.forwardsR#7BD$5b48<8X8X$&&|4^b373R3RT))$/X\ (4OA|#$58H$H!.7.CilO**t}} $ A A ))!"#%! !-!"#%#) ! *!,M  &9]1=MrErNrCrLrMrNs@r.rrsAj(."<"B B r0rcb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $)ri2c>[TU]5 [R"URUR 5Ul[UR[5(a[URUl gURUl grE) rrr%rr(intermediate_sizerr hidden_actstrr intermediate_act_fnrps r.rLayoutLMv3Intermediate.__init__3s` YYv1163K3KL f'' - -'-f.?.?'@D $'-'8'8D $r0rrcJURU5nURU5nU$rErrm)r*rs r.rCLayoutLMv3Intermediate.forward;s& =1 00? r0rprrNs@r.rr2s(9U\\ellr0rcz^\rSrSrU4SjrS\R S\R S\R 4SjrSrU=r $)riBc(>[TU]5 [R"URUR 5Ul[R"UR URS9Ul[R"UR5Ul gr) rrr%rrjr(rr^r_r`rarbrps r.rLayoutLMv3Output.__init__Cs` YYv779K9KL f&8&8f>S>STzz&"<"<= r0rrrcpURU5nURU5nURX-5nU$rErrs r.rCLayoutLMv3Output.forwardIrr0rrrNs@r.rrBrr0rc^\rSrSrU4SjrSrSrSrSSjrSr Sr \ SS \ \ RS \ \ RS \ \ RS \ \ RS \ \ RS\ \ RS\ \ RS\ \ RS\ \S\ \S\ \S\\\44Sjj5rSrU=r$)riPc>[TU]U5 XlUR(a[ U5UlUR (Ga[U5Ul[URUR- 5n[R"[R"SSUR 55Ul[R"[R"SX"-S-UR 55Ul[R&"SS9Ul[R*"UR UR,S9Ul[R&"UR.5UlURR2(dURR4(aUR7X"4S9 [R*"UR SS9Ul[;U5UlUR?5 g)Nrr)prT)r,gư>) rrr+ text_embedrPr@rr patch_embedrrr#r% Parameterrdrr(rrr`pos_dropr^r_rarbrrinit_visual_bboxnormrencoder init_weights)r*r+r5r-s r.rLayoutLMv3Model.__init__RsL     6v>DO     9@D v((6+<+<<=D\\%++aF V%7%7TBDI(0  r0c.URR$rEr@r[)r*s r.get_input_embeddings$LayoutLMv3Model.get_input_embeddingsos...r0c$XRlgrEr)r*rs r.set_input_embeddings$LayoutLMv3Model.set_input_embeddingsrs */'r0cUR5H7up#URRURR U5 M9 g)z Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base class PreTrainedModel N)itemsrr!r prune_heads)r*heads_to_pruner!headss r. _prune_headsLayoutLMv3Model._prune_headsus< +002LE LL  u % / / ; ;E B3r0c [R"[R"SX!SS--U5USSS9n[R"[R"SX!SS--U5USSS9n[R"USSR USS5USSR USS5R SS5USSR USS5USSR USS5R SS5/SS9R SS5n[R"SSUS- US- //5n[R"Xe/SS9Ul g) z: Create the bounding boxes for the visual (patch) tokens. rrtrunc) rounding_modeNr2rs) rddivrestackrepeatr=r7tensorrw visual_bbox)r*r,max_len visual_bbox_x visual_bbox_yr cls_token_boxs r.r~ LayoutLMv3Model.init_visual_bbox}sX LLG!}q'897 CZPQ]bi  LLG!}q'897 CZPQ]bi kkcr"))*Q-;cr"))*Q-;EEaKab!((A:ab!((A:DDQJ     $r1+  ueWq['A+&N%OP  99m%AqIr0cURRUSS5nURU5RU5nU$r )rrrtype)r*rr batch_sizers r.calculate_visual_bbox%LayoutLMv3Model.calculate_visual_bboxs;&&--j!Q? !nnV,11%8 r0c.URU5nUR5up4nURRUSS5n[R "Xb4SS9nUR bX R -nURU5nURU5nU$)Nr2rrs) r{r5rrgrdrwrr}r)r*r>r@rseq_lenr* cls_tokenss r. forward_imageLayoutLMv3Model.forward_images%%l3 ",!2 Q^^**:r2> YY 7Q?  >> %#nn4J]]:. YYz* r0rrxrrrVrrr>rr^r_rc  U bU OURRn U bU OURRn U bU OURRn Ub!UR 5n U upUR nOMUb$UR 5SSn U upUR nO&Ub[ U5n UR nO [S5eUcUbUc[R"U W4US9nUc$[R"W [RUS9nUc:[R"[[W 5S/-5[RUS9nURUUUUUS9nS=nnS=nnUGb*[UR SURR"- 5[UR S URR"- 5nnUR%U5n[R"U UR S 4[RUS9nUb[R&"UU/S S 9nOUnURR((dURR*(aURR*(a<UR-U[RU S 9nUb[R&"UU/S S 9nOUn[R."S UR S [RUS9R1U S 5nUcUbQ[R."S W S US9R3S 5nUR5U 5n[R&"UU/S S 9nOUnUcUb[R&"WU/S S 9nOUnUR7U5nUR9U5nOURR((dURR*(amURR*(aUnURR((a5URR:SS2SW S 24nUR=U5nUnUR?USUWR@S9nURCX`RRD5nURGUUUUUU U U UUS9 nUS nU (d U4US S-$[IUURJURLS9$)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", trust_remote_code=True) >>> 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 ```Nr2zEYou have to specify either input_ids or inputs_embeds or pixel_values)rrr)rrxrVrrr3r rrs)rrr)r) rxrVrrrr^r_rArBrV)'r+rr^use_return_dictr5rlenrrdonesrrr]listr@rr9r#rrwrrrrerrrgr^rbrV expand_asget_extended_attention_maskr get_head_maskr rr rrX)r*rrxrrrVrrr>rr^r_rr seq_lengthrembedding_output final_bboxfinal_position_idsrArBvisual_embeddingsvisual_attention_maskrvisual_position_idsextended_attention_maskencoder_outputssequence_outputs r.rCLayoutLMv3Model.forwards\2C1N-TXT_T_TqTq$8$D $++JjJj &1%&#Q#[#[\]#^L#/#6#6{#CL).LBU3V\])^&)<&$ (A#(99.>@Q-RXY#Z #4 #~~.>? #||,<=  [[ 4 4 8^8^{{55! {{66#;;A?OQ?OC 'tnC d^C  uo% &C C r0rc6^\rSrSrSrSU4SjjrSrSrU=r$)LayoutLMv3ClassificationHeadiqzT Head for sentence-level classification tasks. Reference: RobertaClassificationHead c>[TU]5 X lU(a4[R"UR S-UR 5UlO0[R"UR UR 5UlURb URO URn[R"U5Ul [R"UR UR5Ul g)Nr ) rr pool_featurer%rr(rclassifier_dropoutrar`rb num_labelsout_proj)r*r+rrr-s r.r%LayoutLMv3ClassificationHead.__init__vs ( 6#5#5#96;M;MNDJ6#5#5v7I7IJDJ)/)B)B)NF % %TZTnTn zz"45  &"4"4f6G6GH r0cURU5nURU5n[R"U5nURU5nUR U5nU$rE)rbrrdtanhr)r*rs r.rC$LayoutLMv3ClassificationHead.forwardsI LLO JJqM JJqM LLO MM! r0)rrbrr)FrFrNs@r.rrqs Ir0ra LayoutLMv3 Model with a token classification head on top (a linear layer on top of the final hidden states) e.g. for sequence labeling (information extraction) tasks such as [FUNSD](https://guillaumejaume.github.io/FUNSD/), [SROIE](https://rrc.cvc.uab.es/?ch=13), [CORD](https://github.com/clovaai/cord) and [Kleister-NDA](https://github.com/applicaai/kleister-nda). ) custom_introc^\rSrSrU4Sjr\SS\\RS\\RS\\RS\\RS\\RS\\RS \\RS \\RS \\ S \\ S \\ S\\RS\ \ \ 44Sjj5rSrU=r$) LayoutLMv3ForTokenClassificationicp>[TU]U5 URUl[U5Ul[ R "UR5UlURS:a1[ R"URUR5Ul O[USS9Ul UR5 g)N Fr)rrrrrr%r`rarbrr( classifierrrrps r.r)LayoutLMv3ForTokenClassification.__init__s   ++)&1zz&"<"<=   r ! ii(:(:Frc U bU OURRn URUUUUUUUU U U U S9 n UbUR5nOUR5SSnUSnU SSS2SU24nUR U5nUR U5nSnUb<[ 5nU"URSUR5URS55nU (dU4U SS-nUbU4U-$U$[UUU RU RS9$)ay bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`. Examples: ```python >>> from transformers import AutoProcessor, AutoModelForTokenClassification >>> from datasets import load_dataset >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = AutoModelForTokenClassification.from_pretrained("microsoft/layoutlmv3-base", num_labels=7) >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train", trust_remote_code=True) >>> example = dataset[0] >>> image = example["image"] >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> word_labels = example["ner_tags"] >>> encoding = processor(image, words, boxes=boxes, word_labels=word_labels, return_tensors="pt") >>> outputs = model(**encoding) >>> loss = outputs.loss >>> logits = outputs.logits ```N) rxrrrVrrrr^r_r>r2rrlosslogitsrrX) r+rrr5rbrrr7rrrrX)r*rrxrrrVrrrrr^r_r>rrrrrrloss_fctrs r.rC(LayoutLMv3ForTokenClassification.forwards:^&1%V 'tnV d^V u//0V  u++ ,V V r0rc ^\rSrSrU4Sjr\SS\\RS\\RS\\RS\\RS\\RS\\RS \\RS \\RS \\ S \\ S \\ S\\RS\\RS\ \ \ 44Sjj5rSrU=r$)LayoutLMv3ForQuestionAnsweringic>[TU]U5 URUl[U5Ul[ USS9UlUR5 gNFr)rrrrrr qa_outputsrrps r.r'LayoutLMv3ForQuestionAnswering.__init__sC   ++)&16vER r0rrrrVrrstart_positions end_positionsrr^r_rxr>rc*U bU OURRn URUUUUUUU U U U U S9 nUSnURU5nUR SSS9unnUR S5R 5nUR S5R 5nSnUbUb[UR55S:aUR S5n[UR55S:aUR S5nURS5nURSU5nURSU5n[US9nU"UU5nU"UU5nUU-S- nU (dUU4USS-nUbU4U-$U$[UUUURURS 9$) a_ bbox (`torch.LongTensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. Examples: ```python >>> from transformers import AutoProcessor, AutoModelForQuestionAnswering >>> from datasets import load_dataset >>> import torch >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = AutoModelForQuestionAnswering.from_pretrained("microsoft/layoutlmv3-base") >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train", trust_remote_code=True) >>> example = dataset[0] >>> image = example["image"] >>> question = "what's his name?" >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> encoding = processor(image, question, words, boxes=boxes, return_tensors="pt") >>> start_positions = torch.tensor([1]) >>> end_positions = torch.tensor([3]) >>> outputs = model(**encoding, start_positions=start_positions, end_positions=end_positions) >>> loss = outputs.loss >>> start_scores = outputs.start_logits >>> end_scores = outputs.end_logits ```N rrrVrrrr^r_rxr>rrr2rs) ignore_indexr3)r start_logits end_logitsrrX)r+rrrsplitsqueezerrr5clamprr rrX)r*rrrrVrrrrrr^r_rxr>rrrrr total_loss ignored_indexr start_lossend_lossrs r.rC&LayoutLMv3ForQuestionAnswering.forwardsf&1%d 'tnd d^d u''(d u//0d  u22 3d d r0ra  LayoutLMv3 Model with a sequence classification head on top (a linear layer on top of the final hidden state of the [CLS] token) e.g. for document image classification tasks such as the [RVL-CDIP](https://www.cs.cmu.edu/~aharley/rvl-cdip/) dataset. c^\rSrSrU4Sjr\SS\\RS\\RS\\RS\\RS\\RS\\RS \\RS \\ S \\ S \\ S \\RS\\RS\ \ \ 44Sjj5rSrU=r$)#LayoutLMv3ForSequenceClassificationioc>[TU]U5 URUlXl[ U5Ul[ USS9UlUR5 gr) rrrr+rrrrrrps r.r,LayoutLMv3ForSequenceClassification.__init__wsH   ++ )&16vER r0rrrrVrrrrr^r_rxr>rc HU bU OURRn URUUUUUUUU U U U S9 n U SSS2SSS24nURU5nSnUGbURRcUR S:XaSURlOoUR S:aNUR [R:XdUR [R:XaSURlOSURlURRS:XaI[5nUR S:Xa&U"UR5UR55nOU"X5nOURRS:Xa=[5nU"URSUR 5URS55nO,URRS:Xa[5nU"X5nU (dU4U SS-nUbU4U-$U$[UUU R U R"S 9$) 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", trust_remote_code=True) >>> 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 ```Nrrr regressionsingle_label_classificationmulti_label_classificationr2r)r+rrr problem_typerrrdrrrrrr7rr rrX)r*rrrrVrrrrr^r_rxr>rrrrrrs r.rC+LayoutLMv3ForSequenceClassification.forwards\&1%` 'tn` d^` u''(` u//0`  u.. /` ` r0r)rrrrr)6rKr rtypingrrrrdtorch.nnr%torch.nn.functional functionalr:torch.utils.checkpointrrr activationsr modeling_outputsr r r rmodeling_utilsr pytorch_utilsrutilsrrrconfiguration_layoutlmv3r get_loggerrGloggerModulerrPrrrrr rrrrrrrr__all__rr0r.r s  )) AA! .6 7   H %  Fsryysl...2WbiiWv 299 ")):(bii(VZ  Z | RYY  ryy ] /] ] @ 2996e '@e e Po %>o o dk *Ck k \ r0