[TU]5 URUlSS[R"SURS5UR- -- nUR SU5 g)Nri'g@inv_freq)super__init__dim_bbox_sinusoid_emb_1dr'arangeregister_buffer)selfconfigr3 __class__s r-r5"BrosPositionalEmbedding1D.__init__Ds^ (.(G(G% ell3(E(EsKdNkNkk l  Z2r,pos_seqreturncUR5nUup4nURX4US5URRSSSURS-5-n[R "UR 5UR5/SS9nU$)Nrdim)sizeviewr3r6r'catsincos)r9r=seq_sizeb1b2b3 sinusoid_inppos_embs r-forward!BrosPositionalEmbedding1D.forwardNs{<<> ||BB2T]]5G5G1aQUQnQnrsQs5tt ))\--/1A1A1CD"Mr,)r6 r"r#r$r%r5r'TensorrOr+ __classcell__r;s@r-r/r/As(3u|| r,r/cb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $)BrosPositionalEmbedding2DVc>[TU]5 URUl[U5Ul[U5UlgN)r4r5dim_bboxr/ x_pos_emb y_pos_embr9r:r;s r-r5"BrosPositionalEmbedding2D.__init__Ws1  26:26:r,bboxr>c/n[UR5HXnUS-S:Xa'URURUSU455 M3URUR USU455 MZ [ R "USS9nU$)Nr@r.rArB)rangerZappendr[r\r'rF)r9r_stacki bbox_pos_embs r-rO!BrosPositionalEmbedding2D.forward^swt}}%A1uz T^^DaL9: T^^DaL9: & yyB/ r,)rZr[r\rQrTs@r-rVrVVs(;ELLU\\r,rVcJ^\rSrSrU4SjrS\R 4SjrSrU=r $)BrosBboxEmbeddingsic>[TU]5 [U5Ul[R "UR URSS9Ulg)NF)bias) r4r5rVbbox_sinusoid_embrLineardim_bbox_sinusoid_emb_2ddim_bbox_projectionbbox_projectionr]s r-r5BrosBboxEmbeddings.__init__js= !:6!B!yy)H)H&JdJdkpqr,r_cURSS5nUSSS2SS2SS24USS2SSS2SS24- nURU5nURU5nU$)Nrr) transposerlrp)r9r_bbox_tbbox_posres r-rOBrosBboxEmbeddings.forwardos\1%$1a-(6!T1a-+@@--h7 ++L9 r,)rprlrQrTs@r-rhrhisr ELLr,rhc ^\rSrSrSrU4SjrS S\\RS\\RS\\RS\\RS\R4 S jjr S r U=r $) BrosTextEmbeddingsxzGConstruct the embeddings from word, position and token_type embeddings.c Z>[TU]5 [R"URUR UR S9Ul[R"URUR 5Ul [R"URUR 5Ul [R"UR URS9Ul [R"UR5Ul[#USS5UlUR'S[(R*"UR5R-S55 UR'S[(R."UR0R35[(R4UR0R6S9S S 9 g) N) padding_idxepsposition_embedding_typeabsolute position_ids)rrAtoken_type_idsdtypedeviceF) persistent)r4r5r Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingstype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutgetattrr~r8r'r7expandzerosrrDlongrr]s r-r5BrosTextEmbeddings.__init__{s8 !||F,=,=v?Q?Q_e_r_rs#%<<0N0NPVPbPb#c %'\\&2H2H&J\J\%]"f&8&8f>S>STzz&"<"<= '.v7PR\']$ ^U\\&:X:X-Y-`-`ah-ij   KK!!&&(jj((//    r, input_idsrr inputs_embedsr>c`UbUR5nOUR5SSnUSnUcURSS2SU24nUcv[US5(a-URSS2SU24nUR USU5nUnO8[ R "U[ RURRS9nUcURU5nURU5n XI-n URS:XaURU5n X- n URU 5n URU 5n U $)NrArrrrr)rDrhasattrrrr'rrrrrr~rrr) r9rrrr input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expandedr embeddingsrs r-rOBrosTextEmbeddings.forwards:  #..*K',,.s3K ^  ,,Q  ^ J":  ' ': 5"&":":<"H   -J^^J/ \\*- r,)rrr~rrr)NNNN) r"r#r$r%r&r5rr'rRrOr+rSrTs@r-rxrxxs{Q 4-115/304 #ELL)#!.#u||, #   - #  ##r,rxc$^\rSrSrU4SjrS S\R S\R S\\R S\\R S\\R S\\R S \\R S \\R 4S jjr S r U=r $)BrosSelfAttentionc>[TU]5 URUR-S:wa7[ US5(d&[ SURSURS35eURUl[ URUR- 5UlURUR-Ul[R"URUR5Ul [R"URUR5Ul [R"URUR5Ul [R"UR5Ul[#USS5UlUR$S:XdUR$S :XaGUR&Ul[R("S UR&-S - UR5UlUR,Ulg) Nrembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ()r~r relative_keyrelative_key_queryr@r)r4r5rnum_attention_headsr ValueErrorintattention_head_size all_head_sizerrmquerykeyvaluerattention_probs_dropout_probrrr~rrdistance_embedding is_decoderr]s r-r5BrosSelfAttention.__init__s     : : :a ?PVXhHiHi#F$6$6#78 445Q8  $*#=#= #&v'9'9F 9T=Y=Y]q=q+1+I+ID (&(ll1v7U7U3UXY3Y[_[s[s&tD # ++r,rreattention_mask head_maskencoder_hidden_statesencoder_attention_maskoutput_attentionsr>cURSSURUR4nURU5R U5R SS5n USLn U (acUR U5R U5R SS5n URU5R U5R SS5n UnO`UR U5R U5R SS5n URU5R U5R SS5n [R"XR SS55n URS:XdURS:XGaDUR5Sn[R"U[RURS9R SS5n[R"U[RURS9R SS5nUU- nURUUR -S- 5nUR#U R$S 9nURS:Xa[R&"S U U5nU U-n OHURS:Xa8[R&"S U U5n[R&"S U U5nU U-U-n U RunnnnUR XUU5nUR)/S Q5n[R&"S X45nU U-n U [*R,"UR5- n UbX-n [.R0"SS9"U 5nUR3U5nUbUU-n[R"UU 5nUR)SSSS5R55nUR5SSUR64-nUR"U6nU(aUU4OU4nUR8(aUS-nU$)NrrArr@rrr)rzbhld,lrd->bhlrzbhrd,lrd->bhlr)r@rrrzbnid,bijd->bnijrBrrY)shaperrrrErsrrr'matmulr~rDr7rrrrtoreinsumpermutemathsqrtrSoftmaxr contiguousrr)r9rrerrrrr hidden_shape query_layeris_cross_attention key_layer value_layerattention_scoresrposition_ids_lposition_ids_rdistancepositional_embeddingrelative_position_scoresrelative_position_scores_queryrelative_position_scores_key batch_sizen_headd_headbbox_pos_scoresattention_probs context_layernew_context_layer_shapeoutputss r-rOBrosSelfAttention.forwards&++A.D4L4LdNfNfg jj/44\BLLQPQR 3$> !67<<\JTTUVXYZI**%:;@@NXXYZ\]^K3N/44\BLLQPQRI**]388FPPQRTUVK!<< 5H5HR5PQ  ' '> 9T=Y=Y]q=q&++-a0J"\\*EJJ}OcOcdiijlnopN"\\*EJJ}OcOcdiijkmopN%6H#'#:#:8dFbFb;bef;f#g #7#:#:ARAR#:#S ++~=+0<<8H+Wk+l(#36N#N --1EE16>NP[]q1r./4||,,7  - 9O _kB %--aAq9DDF "/"4"4"6s";t?Q?Q>S"S%**,CD 6G=/2mM] ??'Gr,) rrrrrrrrr~rrNNNNF) r"r#r$r%r5r'rRrr*rOr+rSrTs@r-rrs,826,08<9=49P||PllP!. P ELL) P ( 5 P!) 6P$ELL1P u|| PPr,rcz^\rSrSrU4SjrS\R S\R S\R 4SjrSrU=r $)BrosSelfOutputi&c(>[TU]5 [R"URUR5Ul[R "URURS9Ul[R"UR5Ul gNr|) r4r5rrmrdenserrrrrr]s r-r5BrosSelfOutput.__init__'s` YYv1163E3EF f&8&8f>S>STzz&"<"<= r,r input_tensorr>cpURU5nURU5nURX-5nU$rYrrrr9rrs r-rOBrosSelfOutput.forward-5 =1  ]3 }'CD r,rrrrQrTs@r-rr&6> U\\RWR^R^r,rc^\rSrSrU4SjrSrSS\RS\RS\\RS\\RS\\RS \\RS \\ S \ \R4S jjr S r U=r $) BrosAttentioni4c>[TU]5 [U5Ul[ U5Ul[ 5UlgrY)r4r5rr9routputset pruned_headsr]s r-r5BrosAttention.__init__5s0 %f- $V, Er,c8[U5S:Xag[UURRURRUR 5up[ URRU5URl[ URRU5URl[ URRU5URl [ URRUSS9URl URR[U5- URlURRURR-URl UR RU5Ulg)NrrrB)lenrr9rrrrrrrrrrunion)r9headsindexs r- prune_headsBrosAttention.prune_heads;s u:? 7  II ) ) II ) )     -TYY__eD *499==%@ ,TYY__eD .t{{/@/@%QO )- (E(EE (R %"&))"?"?$))B_B_"_  --33E:r,rrerrrrrr>c lURUUUUUUUS9nURUSU5n U 4USS-n U $)Nrrerrrrrrr)r9r) r9rrerrrrr self_outputsattention_outputrs r-rOBrosAttention.forwardPs\yy'%)"7#9/!  ;;|A F#% QR(88r,)rrr9r)r"r#r$r%r5rr'rRrboolr*rOr+rSrTs@r-rr4s" ;226,08<9=,1||ll!.  ELL)  ( 5 !) 6$D> u|| r,rcb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $)BrosIntermediateiic>[TU]5 [R"URUR 5Ul[UR[5(a[URUl gURUl grY) r4r5rrmrintermediate_sizer isinstance hidden_actstrr intermediate_act_fnr]s r-r5BrosIntermediate.__init__js` YYv1163K3KL f'' - -'-f.?.?'@D $'-'8'8D $r,rr>cJURU5nURU5nU$rYrr)r9rs r-rOBrosIntermediate.forwardrs& =1 00? r,r rQrTs@r-rris(9U\\ellr,rcz^\rSrSrU4SjrS\R S\R S\R 4SjrSrU=r $) BrosOutputixc(>[TU]5 [R"URUR 5Ul[R"UR URS9Ul[R"UR5Ul gr) r4r5rrmrrrrrrrrr]s r-r5BrosOutput.__init__ys` YYv779K9KL f&8&8f>S>STzz&"<"<= r,rrr>cpURU5nURU5nURX-5nU$rYrrs r-rOBrosOutput.forwardrr,rrQrTs@r-r r xrr,r c^\rSrSrU4SjrSS\R S\R S\\RS\\RS\\RS\\RS \\ S \ \R 4S jjr S r S r U=r$) BrosLayericv>[TU]5 URUlSUl[ U5UlUR UlURUlUR(a/UR (d[US35e[ U5Ul [U5Ul [U5Ul g)Nrz> should be used as a decoder model if cross attention is added)r4r5chunk_size_feed_forward seq_len_dimr attentionradd_cross_attention Exceptioncrossattentionr intermediater rr]s r-r5BrosLayer.__init__s '-'E'E$&v. ++#)#=#=  # #??4&(f ghh"/"7D ,V4 ( r,rrerrrrrr>c URUUUUUS9nUSn UR(aUSSn OUSSn UR(aCUb@[US5(a[SUS35eUR U UUUUUS9n U Sn XSS-n [ UR URURU 5n U 4U -n UR(aU S -n U $) N)rerrrrrrArz'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`)rrrrrrY) rrrrrrfeed_forward_chunkrr) r9rrerrrrrself_attention_outputsrrcross_attention_outputs layer_outputs r-rOBrosLayer.forwards%"& %)/ "0" 2!4 ??,Qr2G,QR0G ??4@t-..=dVDde'+&9&9 -#&;'="3 ':' # 7q9 " ==G0  # #  ( (       /G+ ??'Gr,cJURU5nURX!5nU$rY)rr)r9rintermediate_outputr!s r-rBrosLayer.feed_forward_chunks)"//0@A{{#6I r,)rrrrrrrrr)r"r#r$r%r5r'rRrr(rr*rOrr+rSrTs@r-rrs )$7;15=A>B,16||6ll6!!2!23 6 E--. 6 ((9(9: 6!)):): ;6$D>6 u|| 6pr,rc<^\rSrSrU4Sjr\SS\RS\RS\\RS\\RS\\RS\\RS \\ S \\ S \\ S \ \ \R\ 44S jj5rSrU=r$) BrosEncoderic>[TU]5 Xl[R"[ UR 5Vs/sHn[U5PM sn5Ulgs snfrY) r4r5r:r ModuleListranum_hidden_layersrlayer)r9r:_r;s r-r5BrosEncoder.__init__sG  ]]uVE]E]?^#_?^!If$5?^#_` #_sArrerrrrroutput_hidden_states return_dictr>c U(aSOSn U(aSOSn U(aURR(aSOSn [UR5HcupU(aX4-n UbXMOSnU"UUUUUUUS9nUSnU(dM4U US4-n URR(dMZU US4-n Me U(aX4-n [ UU U U S9$)Nr!rrrr@)last_hidden_staterr cross_attentions)r:r enumerater+r )r9rrerrrrrr.r/all_hidden_statesall_self_attentionsall_cross_attentionsrd layer_modulelayer_head_mask layer_outputss r-rOBrosEncoder.forwards#7BD$5b4%64;;;Z;Zr`d(4OA#$58H$H!.7.CilO(+)-)&;'="3M*!,M  &9]1=MB,1/4&*. ||. ll. !!2!23 . E--. . ((9(9: . !)):): ;. $D>. 'tn. d^.  uU\\"$FF G. . r,r'cb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $) BrosPooleri c>[TU]5 [R"URUR5Ul[R "5UlgrY)r4r5rrmrrTanh activationr]s r-r5BrosPooler.__init__ s9 YYv1163E3EF '')r,rr>c\USS2S4nURU5nURU5nU$)Nr)rr?)r9rfirst_token_tensor pooled_outputs r-rOBrosPooler.forwards6+1a40 #56  6 r,)r?rrQrTs@r-r<r< s($ U\\ellr,r<cb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $)BrosRelationExtractoric\>[TU]5 URUlURUlURUlUR Ul[R"UR 5Ul [R"URURUR -5Ul [R"URURUR -5Ul [R"[R"SUR55Ulg)Nr)r4r5 n_relationsrbackbone_hidden_sizehead_hidden_sizeclassifier_dropout_probrrdroprmrr Parameterr'r dummy_noder]s r-r5BrosRelationExtractor.__init__s !--$*$6$6! & 2 2'-'E'E$JJt;;< YYt88$:J:JTMbMb:bc 99T668H8H4K`K`8`a,,u{{1d6O6O'PQr,rrc URURU55nURRS5R SUR S5S5n[ R"X#/SS9nURURU55nURUR S5UR S5URUR5nURUR S5UR S5URUR5n[ R"URSSSS5URSSSS55nU$)Nrraxisr@r)rrLrN unsqueezerepeatrDr'rFrrErHrJrr)r9rr dummy_vecrelation_scores r-rOBrosRelationExtractor.forward*sjj;!78 OO--a0779>>!;LaP IIy41= HHTYYy12 !&&   Q !1!1!!4d6F6FH]H] NN9>>!#4innQ6GIYIY[_[p[pq    1a +Y->->q!Q-J r,)rIrKrLrNrJrrHrrQrTs@r-rFrFs) R5<<ELLr,rFcF\rSrSr%\\S'SrS\R4Sjr Sr g)BrosPreTrainedModeli=r:brosmodulecpURRn[U[R5(aWUR R RSUS9 URb%URR R5 gg[U[R5(adUR R RSUS9 URb2UR R URR5 gg[U[R5(aJURR R5 UR R RS5 g[U[5(a)[RRUR US9 gg)zInitialize the weightsr2)meanstdNg?)r^)r:initializer_rangerrrmweightdatanormal_rkzero_rr{rfill_rFinitrN)r9r[r^s r- _init_weights!BrosPreTrainedModel._init_weightsBs2kk++ fbii ( ( MM   & &CS & 9{{&   &&('  - - MM   & &CS & 9!!- ""6#5#56<<>.  - - KK   " " $ MM   $ $S )  5 6 6 GGOOF--3O 77r,r!N) r"r#r$r%rr)base_model_prefixrModulerfr+r!r,r-rYrY=s 8BII8r,rYc^\rSrSrSU4SjjrSrSrSr\\ SS\ \ RS\ \ RS\ \ RS \ \ RS \ \ RS \ \ RS \ \ RS \ \ RS\ \ RS\ \ S\ \ S\ \ S\\\ R\44Sjj55rSrU=r$) BrosModeliVc>[TU]U5 Xl[U5Ul[ U5Ul[U5UlU(a [U5OSUl UR5 g)z^ add_pooling_layer (bool, *optional*, defaults to `True`): Whether to add a pooling layer N) r4r5r:rxrrhbbox_embeddingsr'encoderr<pooler init_weights)r9r:add_pooling_layerr;s r-r5BrosModel.__init__XsX   ,V41&9"6* ,=j(4  r,c.URR$rYrr)r9s r-get_input_embeddingsBrosModel.get_input_embeddingshs...r,c$XRlgrYrt)r9rs r-set_input_embeddingsBrosModel.set_input_embeddingsks */'r,cUR5H7up#URRURR U5 M9 g)z Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base class PreTrainedModel N)itemsrnr+rr)r9heads_to_pruner+rs r- _prune_headsBrosModel._prune_headsns< +002LE LL  u % / / ; ;E B3r,rr_rrrrrrrrr.r/r>c ^U bU OURRn U bU OURRn U bU OURRn UbUb [ S5eUbUR 5n O"UbUR 5SSn O [ S5eUc [ S5eU upUb UR O UR nUc[R"U US9nUcr[URS5(a3URRSS2SU24nURX5nUnO$[R"U [RUS9nURX=U5nURR (aEUbBUR 5unnnUU4nU c[R"UUS9n UR#U 5nOSnUR%X`RR&5nURUUUUS 9nUR(SS :XaUSS2SS2/S Q4nX RR*-nUR-U5nUR/UUUUUUU U S S 9 nUSnUR0bUR1U5OSn[3UUUR4UR6UR8S9$)aM bbox ('torch.FloatTensor' of shape '(batch_size, num_boxes, 4)'): Bounding box coordinates for each token in the input sequence. Each bounding box is a list of four values (x1, y1, x2, y2), where (x1, y1) is the top left corner, and (x2, y2) is the bottom right corner of the bounding box. Examples: ```python >>> import torch >>> from transformers import BrosProcessor, BrosModel >>> processor = BrosProcessor.from_pretrained("jinho8345/bros-base-uncased") >>> model = BrosModel.from_pretrained("jinho8345/bros-base-uncased") >>> encoding = processor("Hello, my dog is cute", add_special_tokens=False, return_tensors="pt") >>> bbox = torch.tensor([[[0, 0, 1, 1]]]).repeat(1, encoding["input_ids"].shape[-1], 1) >>> encoding["bbox"] = bbox >>> outputs = model(**encoding) >>> last_hidden_states = outputs.last_hidden_state ```NzDYou cannot specify both input_ids and inputs_embeds at the same timerAz5You have to specify either input_ids or inputs_embedszYou have to specify bbox)rrr)rrrr)rrr@rr@rrrT)rerrrrrr.r/r)r1 pooler_outputrr r2)r:rr.use_return_dictrrDrr'onesrrrrrrget_extended_attention_maskrinvert_attention_mask get_head_maskr*r bbox_scalermrnror rr r2)r9rr_rrrrrrrrr.r/rrrrrrextended_attention_maskencoder_batch_sizeencoder_sequence_lengthr,encoder_hidden_shapeencoder_extended_attention_maskembedding_output scaled_bboxbbox_position_embeddingsencoder_outputssequence_outputrCs r-rOBrosModel.forwardvsP2C1N-TXT_T_TqTq$8$D $++JjJj &1%cd d  "#..*K  &',,.s3KTU U <78 8!, %.%:!! @T@T  !"ZZ FCN  !t(899*.//*H*HKZK*X'3J3Q3QR\3i0!A!&[ SY!Z150P0PQ_nt0u ;; ! !&;&G=R=W=W=Y :  7$68O#P %-).4HQW)X&.2.H.HI_.` +.2 +&&y++2O2OP ??%)' +  ::b>Q 1667D[[333 #'#7#7 #D ,, 12"7#B/!5'  *!,8< 8O O4UY ;-')77&11,==   r,)rmr:rrnro)T NNNNNNNNNNNN)r"r#r$r%r5rurxr}rrrr'rRrrr*r rOr+rSrTs@r-rkrkVsZ /0C-1'+1515/3,0048<9=,0/3&*} ELL)} u||$} !. } !. } u||, } ELL)}   -}  ( 5} !) 6} $D>} 'tn} d^}  uU\\"$PP Q} } r,rkc^\rSrSrS/rU4Sjr\\SS\\ RS\\ RS\\ RS\\ RS\\ RS \\ RS \\ RS \\ RS \\ RS \\ S\\ S\\ S\ \ \ R\44Sjj55rSrU=r$)BrosForTokenClassificationirocn>[TU]U5 URUl[U5Ul[ US5(a UR O URn[R"U5Ul [R"URUR5Ul UR5 gNclassifier_dropout)r4r5 num_labelsrkrZrrrrrrrmr classifierrpr9r:rr;s r-r5#BrosForTokenClassification.__init__s   ++f% )09M)N)NF % %TZTnTn zz"45 ))F$6$68I8IJ r,rr_rbbox_first_token_maskrrrrlabelsrr.r/r>c U bU OURRn URUUUUUUUU U SS9 n U SnURU5nUR U5nSnU b[ 5nUbJUR S5nU"UR SUR5UU R S5U5nO2U"UR SUR5U R S55n[UUU RU RS9$)aP bbox ('torch.FloatTensor' of shape '(batch_size, num_boxes, 4)'): Bounding box coordinates for each token in the input sequence. Each bounding box is a list of four values (x1, y1, x2, y2), where (x1, y1) is the top left corner, and (x2, y2) is the bottom right corner of the bounding box. bbox_first_token_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to indicate the first token of each bounding box. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. Examples: ```python >>> import torch >>> from transformers import BrosProcessor, BrosForTokenClassification >>> processor = BrosProcessor.from_pretrained("jinho8345/bros-base-uncased") >>> model = BrosForTokenClassification.from_pretrained("jinho8345/bros-base-uncased") >>> encoding = processor("Hello, my dog is cute", add_special_tokens=False, return_tensors="pt") >>> bbox = torch.tensor([[[0, 0, 1, 1]]]).repeat(1, encoding["input_ids"].shape[-1], 1) >>> encoding["bbox"] = bbox >>> outputs = model(**encoding) ```NT) r_rrrrrrr.r/rrArlogitsrr ) r:rrZrrrrErr rr )r9rr_rrrrrrrrr.r/rrrrloss_fcts r-rO"BrosForTokenClassification.forward sZ&1%O 'tnO d^O  uU\\"$99 :O O r,ra Bros Model with a token classification head on top (initial_token_layers and subsequent_token_layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. The initial_token_classifier is used to predict the first token of each entity, and the subsequent_token_classifier is used to predict the subsequent tokens within an entity. Compared to BrosForTokenClassification, this model is more robust to serialization errors since it predicts next token from one token. c!^\rSrSrS/rU4Sjr\\SS\\ RS\\ RS\\ RS\\ RS\\ RS \\ RS \\ RS \\ RS \\ RS \\ RS\\ S\\ S\\ S\ \ \ R\44Sjj55rSrU=r$)!BrosSpadeEEForTokenClassificationi]roc x>[TU]U5 XlURUlURUlUR Ul[U5Ul[US5(a URO URn[R"[R"U5[R"UR UR 5[R"U5[R"UR UR55Ul[#U5UlUR'5 gr)r4r5r:rrHrrIrkrZrrrr Sequentialrrminitial_token_classifierrFsubsequent_token_classifierrprs r-r5*BrosSpadeEEForTokenClassification.__init__is    ++!--$*$6$6!f% )09M)N)NF % %TZTnTn  )+ JJ) * IIf((&*<*< = JJ) * IIf((&*;*; < ) %,A+H( r,rr_rrrrrrinitial_token_labelssubsequent_token_labelsrr.r/r>c U bU OURRn URUUUUUUUU U SS9 nUSnURSS5R 5nUR U5RSS5R 5nUR X5RS5nSU- nURunnURn[R"U[R"US/5RU5/SS9R5nURUSS2SSS24[R "UR"5R$5n[R&"UUS-5RU[RS9nURUSSS2SS24[R "UR"5R$5nUR)S5R5nSnU bU b[+5nU R)S5n Ub:UR)S5nU"UR)SUR,5UX5nO#U"UR)SUR,5U 5nU R)S5n U"UR)SUS-5UU U5nUU-n[/UUUUR0UR2S 9$) a bbox ('torch.FloatTensor' of shape '(batch_size, num_boxes, 4)'): Bounding box coordinates for each token in the input sequence. Each bounding box is a list of four values (x1, y1, x2, y2), where (x1, y1) is the top left corner, and (x2, y2) is the bottom right corner of the bounding box. bbox_first_token_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to indicate the first token of each bounding box. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. initial_token_labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for the initial token classification. subsequent_token_labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for the subsequent token classification. Examples: ```python >>> import torch >>> from transformers import BrosProcessor, BrosSpadeEEForTokenClassification >>> processor = BrosProcessor.from_pretrained("jinho8345/bros-base-uncased") >>> model = BrosSpadeEEForTokenClassification.from_pretrained("jinho8345/bros-base-uncased") >>> encoding = processor("Hello, my dog is cute", add_special_tokens=False, return_tensors="pt") >>> bbox = torch.tensor([[[0, 0, 1, 1]]]).repeat(1, encoding["input_ids"].shape[-1], 1) >>> encoding["bbox"] = bbox >>> outputs = model(**encoding) ```NT rr_rrrrrrr.r/rrrQrrrA)rrrrr )r:rrZrsrrrsqueezerrr'rFrrr masked_fillfinformineyerErrrrr )r9rr_rrrrrrrrrr.r/rlast_hidden_statesrrinv_attention_maskrmax_seq_lengthrinvalid_token_maskself_token_masksubsequent_token_maskrrinitial_token_losssubsequent_token_losss r-rO)BrosSpadeEEForTokenClassification.forwardsd&1%  +0G0S')H$8#<#% ! &(==D!5$;!//))   r,)rIrZr:rrHrr) NNNNNNNNNNNNN)r"r#r$r%rr5rrrr'rRrrr*rrOr+rSrTs@r-rr]sj+4&2-1'+158<15/3,0047;:>,0/3&*o ELL)o u||$o !. o ( 5 o !. o u||,o ELL)o   -o 'u||4o "*%,,!7o $D>o 'tno d^o  uU\\"O3 4o o r,rz Bros Model with a token classification head on top (a entity_linker layer on top of the hidden-states output) e.g. for Entity-Linking. The entity_linker is used to predict intra-entity links (one entity to another entity). c^\rSrSrS/rU4Sjr\\SS\\ RS\\ RS\\ RS\\ RS\\ RS \\ RS \\ RS \\ RS \\ RS \\ S\\ S\\ S\ \ \ R\44Sjj55rSrU=r$)!BrosSpadeELForTokenClassificationirocH>[TU]U5 XlURUlURUlUR Ul[U5Ul[US5(a URO UR [U5Ul UR5 gr)r4r5r:rrHrrIrkrZrrrrF entity_linkerrpr]s r-r5*BrosSpadeELForTokenClassification.__init__s    ++!--$*$6$6!f% &-f6J&K&K " "QWQkQk26: r,rr_rrrrrrrrr.r/r>c U bU OURRn URUUUUUUUU U SS9 n U SnURSS5R 5nUR X5R S5nSnU Gba[5nURunnURn[R"UUS-5RU[RS9nURS5n[R"U)[R "US/[RUS9/SS 9nUR#USS2SSS24[R$"UR&5R(5nUR#USSS2SS24[R$"UR&5R(5nU"URSUS-5UU RS5U5n[+UUU R,U R.S 9$) a^ bbox ('torch.FloatTensor' of shape '(batch_size, num_boxes, 4)'): Bounding box coordinates for each token in the input sequence. Each bounding box is a list of four values (x1, y1, x2, y2), where (x1, y1) is the top left corner, and (x2, y2) is the bottom right corner of the bounding box. bbox_first_token_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to indicate the first token of each bounding box. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. Examples: ```python >>> import torch >>> from transformers import BrosProcessor, BrosSpadeELForTokenClassification >>> processor = BrosProcessor.from_pretrained("jinho8345/bros-base-uncased") >>> model = BrosSpadeELForTokenClassification.from_pretrained("jinho8345/bros-base-uncased") >>> encoding = processor("Hello, my dog is cute", add_special_tokens=False, return_tensors="pt") >>> bbox = torch.tensor([[[0, 0, 1, 1]]]).repeat(1, encoding["input_ids"].shape[-1], 1) >>> encoding["bbox"] = bbox >>> outputs = model(**encoding) ```NTrrrrrArrQr)r:rrZrsrrrrrrr'rrrrErFrrrrrr rr )r9rr_rrrrrrrrr.r/rrrrrrrrrmasks r-rO)BrosSpadeELForTokenClassification.forward sX&1%?Ev{{SUW[G\]D$!//))   r,)rIrZr:rrHrrrrTs@r-rrsT+4& -1'+158<15/3,004)-,0/3&*Y ELL)Y u||$Y !. Y ( 5 Y !. Y u||,Y ELL)Y   -Y &Y $D>Y 'tnY d^Y  uU\\"$99 :Y Y r,r)rYrkrrr)9r&r dataclassesrtypingrrr'torch.utils.checkpointrtorch.nnr activationsr modeling_layersr modeling_outputsr r r modeling_utilsr pytorch_utilsrrrutilsrrrrconfiguration_brosr get_loggerr"loggerrrir/rVrhrxrrrrr rr'r<rFrYrkrrr__all__r!r,r-rs !" %!9 .llKK*   H %  :k:  :" * &  >>Bi iZ RYY 1BII1j ryy   J*JZ5 "))5 r  BIID8/880^ #^ ^ Ba !4a a HM (;M M ` l (;l  l ^ r,