cCi4SrSSKJr SSKrSSKrSSKrSSKrSSK J r SSK J r J r JrJrJrJr SSKJrJrJrJrJrJrJrJrJrJr SSKJrJrJ r SS K!J"r"J#r#J$r$J%r% S S K&J'r' \$RP"\)5r*S r+"S S\RXRZ5r."SS\RXRZ5r/"SS\RXRZ5r0"SS\RXRZ5r1"SS\RXRZ5r2"SS\RXRZ5r3"SS\RXRZ5r4"SS\RXRZ5r5"SS\RXRZ5r6"SS \RXRZ5r7"S!S"\RXRZ5r8"S#S$\RXRZ5r9\"S%S&\RXRZ55r:"S'S(\5r;S)r\""S.\<5"S/S0\;\55r?\""S1\<5"S2S3\;\55r@\""S4\<5"S5S6\;\55rA\""S7\<5"S8S9\;\55rB/S:QrCg);zTF 2.0 LayoutLM model.) annotationsN)get_tf_activation)+TFBaseModelOutputWithPastAndCrossAttentions.TFBaseModelOutputWithPoolingAndCrossAttentionsTFMaskedLMOutputTFQuestionAnsweringModelOutputTFSequenceClassifierOutputTFTokenClassifierOutput) TFMaskedLanguageModelingLossTFModelInputTypeTFPreTrainedModelTFQuestionAnsweringLossTFSequenceClassificationLossTFTokenClassificationLossget_initializerkeraskeras_serializable unpack_inputs)check_embeddings_within_bounds shape_liststable_softmax)add_start_docstrings%add_start_docstrings_to_model_forwardloggingreplace_return_docstrings)LayoutLMConfigrcn^\rSrSrSrSU4SjjrSSjrS S SjjrSrU=r $) TFLayoutLMEmbeddings8zGConstruct the embeddings from word, position and token_type embeddings.c l>[TU]"S0UD6 XlURUlURUlUR UlUR Ul[RRURSS9Ul [RRURS9Ulg)N LayerNormepsilonnamerate)super__init__config hidden_sizemax_position_embeddingsmax_2d_position_embeddingsinitializer_rangerlayersLayerNormalizationlayer_norm_epsr#Dropouthidden_dropout_probdropoutselfr,kwargs __class__s k/home/james-whalen/.local/lib/python3.13/site-packages/transformers/models/layoutlm/modeling_tf_layoutlm.pyr+TFLayoutLMEmbeddings.__init__;s "6" !--'-'E'E$*0*K*K'!'!9!988AVAV]h8i||++1K1K+L c[R"S5 URSURRUR /[ UR5S9UlSSS5 [R"S5 URSURRUR /[ UR5S9Ul SSS5 [R"S5 URSURUR /[ UR5S9Ul SSS5 [R"S5 URSURUR /[ UR5S9UlSSS5 [R"S5 URSURUR /[ UR5S9UlSSS5 [R"S 5 URSURUR /[ UR5S9UlSSS5 [R"S 5 URSURUR /[ UR5S9UlSSS5 UR$(agS Ul['US S5bf[R"UR(R*5 UR(R-SSURR /5 SSS5 gg!,(df  GN=f!,(df  GN=f!,(df  GN9=f!,(df  GN=f!,(df  GN=f!,(df  GNU=f!,(df  GN =f!,(df  g=f) Nword_embeddingsweight)r&shape initializertoken_type_embeddings embeddingsposition_embeddingsx_position_embeddingsy_position_embeddingsh_position_embeddingsw_position_embeddingsTr#)tf name_scope add_weightr, vocab_sizer-rr0r@type_vocab_sizerCr.rEr/rFrGrHrIbuiltgetattrr#r&buildr8 input_shapes r;rQTFLayoutLMEmbeddings.buildFs ]], -//{{--t/?/?@+D,B,BC*DK.]]2 3)-!{{22D4D4DE+D,B,BC*9*D &4]]0 1'+!33T5E5EF+D,B,BC(7(D $2]]2 3)-!668H8HI+D,B,BC*9*D &4]]2 3)-!668H8HI+D,B,BC*9*D &4]]2 3)-!668H8HI+D,B,BC*9*D &4]]2 3)-!668H8HI+D,B,BC*9*D &4 ::  4d + 7t~~223$$dD$++2I2I%JK43 8i. -4 32 14 34 34 34 343sfA L5?A M'AMAM+#AM=ANAN!83N35 M M M(+ M:= N  N! N03 Oc8UcUceUb>[XRR5 [R"UR US9n[ U5SSnUc[R"USS9nUc+[R"[R"SUSS9SS9nUc+[R"[R"SUSS9SS9nUc[R"US/-SS 9n[R"URUSS2SS2S45n[R"URUSS2SS2S 45n [R"URUSS2SS2S 45n [R"URUSS2SS2S 45n [R"URUSS2SS2S 4USS2SS2S 4- 5n [R"URUSS2SS2S 4USS2SS2S4- 5n[R"URUS9n[R"UR US9nUU-U-U-U -U -U -U -U-nUR#US9nUR%UUS9nU$![an [S 5U eSn A ff=f)zr Applies embedding based on inputs tensor. Returns: final_embeddings (`tf.Tensor`): output embedding tensor. N)paramsindicesrdimsvalue)startlimitaxis)r[rrz:The `bbox`coordinate values should be within 0-1000 range.inputsrctraining)rr,rMrJgatherr@rfill expand_dimsrangerFrG IndexErrorrHrIrErCr#r6)r8 input_idsbbox position_idstoken_type_ids inputs_embedsrerSleft_position_embeddingsupper_position_embeddingsright_position_embeddingslower_position_embeddingserHrIposition_embedstoken_type_embedsfinal_embeddingss r;callTFLayoutLMEmbeddings.calls%-*?@@  *9kk6L6L MIIT[[)LM /4  !WW+Q?N  >>"((+b/*RYZ  >>"((+b/*RYZ <77;!,A6D b')yy1K1KTRSUVXYRY]'[ $(* $2L2LdSTVWYZSZm(\ %(* $2L2LdSTVWYZSZm(\ %(* $2L2LdSTVWYZSZm(\ %!# $*D*Dd1aQR7mVZ[\^_ab[bVcFc d " $*D*Dd1aQR7mVZ[\^_ab[bVcFc d))4+C+C\ZIIT-G-GQ_`   ' '(  ( (  ( (  ($ $$ $  >>1A>B<> J JJ)r#rOr,r6rHr-r0r/r.rErCrIr@rFrGr,rN)NNNNNF)rktf.Tensor | Nonerlr|rmr|rnr|ror|reboolreturn tf.Tensor) __name__ __module__ __qualname____firstlineno____doc__r+rQrx__static_attributes__ __classcell__r:s@r;r r 8suQ M7Lv'+!%)-+/*.< #< < ' < ) < ( < <  < < r=r cr^\rSrSrSU4SjjrSSjrS S SjjrS SjrSrU=r $) TFLayoutLMSelfAttentionc >[TU]"S 0UD6 URUR-S:wa&[ SURSURS35eURUl[ URUR- 5UlURUR -Ul[R"UR 5Ul [RRUR[UR5SS9Ul[RRUR[UR5SS9Ul[RRUR[UR5SS9Ul[RR'UR(S 9UlUR,UlXlg) NrzThe hidden size (z6) is not a multiple of the number of attention heads ()queryunitskernel_initializerr&keyr[r'r))r*r+r-num_attention_heads ValueErrorintattention_head_size all_head_sizemathsqrtsqrt_att_head_sizerr1Denserr0rrr[r4attention_probs_dropout_probr6 is_decoderr,r7s r;r+ TFLayoutLMSelfAttention.__init__s "6"    : : :a ?#F$6$6#78''-'A'A&B!E  $*#=#= #&v'9'9FURURUS9U 5n URUR US9U 5n URX5nUR(aX4n[ R"XSS9n[ R"URURS9n[ R"UU5nUb[ R"X5n[US S 9nURUUS 9nUb[ R "UU5n[ R"UU 5n[ R""U/S QS 9n[ R$"UU S UR&4S9nU(aUU4OU4nUR(aUU4-nU$)Nrrbrrar^T) transpose_bdtyperX)logitsr_rdrrr)rrrrr[rJconcatrmatmulcastrrdivideaddrr6multiplyrrr)r8 hidden_statesattention_mask head_maskencoder_hidden_statesencoder_attention_maskpast_key_valueoutput_attentionsrermixed_query_layeris_cross_attention key_layer value_layer query_layerattention_scoresdkattention_probsattention_outputoutputss r;rxTFLayoutLMSelfAttention.callsq .q1  JJmJ< 3$> ."<&q)I(+K3N 11$((BW(2XZdeI33DJJF[J4\^hiK3N  '11$((-(2PR\]I33DJJmJ4TV`aK >!#4i"@qII))^A%6 $D1MK11$((-(2PR\]I33DJJmJ4TV`aK//0AN ??(5N99[N WWT,,4D4J4J K99%5r:  %!vv&6G )0@rJ,,o,Q   kk/9EO99_kB<<(8|L::-=jRTVZVhVhEij9J#_5QaPc ?? 11Gr=cPUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  N=f!,(df  g=f)NTrrr[) rOrPrJrKrr&rQr,r-rr[rRs r;rQTFLayoutLMSelfAttention.build4s ::  4$ ' 3tzz/   $dkk.E.E!FG0 4 % 1txx}}-dDKK,C,CDE. 4$ ' 3tzz/   $dkk.E.E!FG0/ 4 0/.-0/s$3E53F83F5 F F F%) rrrOr,r6rrrrrr[rz)rrrrr~rF)rrrrrrrrrrrtuple[tf.Tensor]rr}rer}r~rr{) rrrrr+rrxrQrrrs@r;rrs87 O O"O O ) O !* O)O OO Ob H Hr=rcD^\rSrSrSU4SjjrSSSjjrS SjrSrU=r$) TFLayoutLMSelfOutputiDc f>[TU]"S0UD6 [RR UR [ UR5SS9Ul[RRURSS9Ul [RRURS9UlXlgNdenserr#r$r'r)r*r+rr1rr-rr0rr2r3r#r4r5r6r,r7s r;r+TFLayoutLMSelfOutput.__init__E "6"\\''$$IaIa9bip(  88AVAV]h8i||++1K1K+L  r=cdURUS9nURXS9nURX-S9nU$Nrbrdrr6r#r8r input_tensorres r;rxTFLayoutLMSelfOutput.callO; - 8  M M m.JK r=cHUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=fNTrr# rOrPrJrKrr&rQr,r-r#rRs r;rQTFLayoutLMSelfOutput.buildV ::  4$ ' 3tzz/   $dkk.E.E!FG0 4d + 7t~~223$$dD$++2I2I%JK43 80/433D3D D D!r#rOr,rr6rzrrrrrrer}r~rr{ rrrrr+rxrQrrrs@r;rrD L Lr=rcn^\rSrSrSU4SjjrSrSS SjjrS SjrSrU=r $) TFLayoutLMAttentionicc b>[TU]"S0UD6 [USS9Ul[ USS9Ulg)Nr8r&outputr))r*r+rself_attentionr dense_outputr7s r;r+TFLayoutLMAttention.__init__ds1 "6"5f6J0hGr=c[er{NotImplementedError)r8headss r; prune_headsTFLayoutLMAttention.prune_headsjs!!r=c jURUUUUUUUUS9n URU SXS9n U 4U SS-n U $)Nrrrrrrrrerrrrer)rr) r8rrrrrrrre self_outputsrrs r;rxTFLayoutLMAttention.callmsm**&)"7#9)/+   ,,&q/ - $% QR(88r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTrr)rOrPrJrKrr&rQrrRs r;rQTFLayoutLMAttention.builds ::  4)4 0 <t22778##))$/9 4 . :t00556!!''-76 ;9876C.C% C"% C3)rOrrrzr)rrrrrrrrrrrrrr}rer}r~rr{) rrrrr+rrxrQrrrs@r;rrcszH ""  )  !* )  : . .r=rc@^\rSrSrSU4SjjrSSjrSSjrSrU=r$) TFLayoutLMIntermediateic J>[TU]"S0UD6 [RR UR [ UR5SS9Ul[UR[5(a[UR5Ul OURUl Xlg)Nrrr))r*r+rr1rintermediate_sizerr0r isinstance hidden_actstrrintermediate_act_fnr,r7s r;r+TFLayoutLMIntermediate.__init__s "6"\\''**vOgOg?hov(   f'' - -'89J9J'KD $'-'8'8D $ r=cFURUS9nURU5nU$Nrb)rrr8rs r;rxTFLayoutLMIntermediate.calls( - 8 00? r=c@UR(agSUl[USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  g=fNTr rOrPrJrKrr&rQr,r-rRs r;rQTFLayoutLMIntermediate.buildm ::  4$ ' 3tzz/   $dkk.E.E!FG0/ 4// 3B B)rOr,rrrzrrr~rr{rrs@r;rrs  HHr=rcD^\rSrSrSU4SjjrSSSjjrS SjrSrU=r$) TFLayoutLMOutputic f>[TU]"S0UD6 [RR UR [ UR5SS9Ul[RRURSS9Ul [RRURS9UlXlgrrr7s r;r+TFLayoutLMOutput.__init__rr=cdURUS9nURXS9nURX-S9nU$rrrs r;rxTFLayoutLMOutput.callrr=cHUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=fr) rOrPrJrKrr&rQr,rr#r-rRs r;rQTFLayoutLMOutput.builds ::  4$ ' 3tzz/   $dkk.K.K!LM0 4d + 7t~~223$$dD$++2I2I%JK43 80/43rrrzrrr{rrs@r;rrrr=rch^\rSrSrSU4SjjrSSSjjrS SjrSrU=r$) TFLayoutLMLayeric B>[TU]"S0UD6 [USS9UlURUlUR UlUR (a.UR(d[ US35e[USS9Ul[USS9Ul [USS9Ul g)N attentionrz> should be used as a decoder model if cross attention is addedcrossattention intermediaterr)) r*r+rrradd_cross_attentionrrrrr bert_outputr7s r;r+TFLayoutLMLayer.__init__s "6",V+F ++#)#=#=  # #?? D6)g!hii"5fCS"TD 26O+FBr=c UbUSSOSn URUUUSSU UUS9n U Sn UR(a U SSn U Sn OU SSn SnUR(aZUbW[US5(d[SUS35eUbUS SOSnUR U UUUUUUUS9nUSn U USS-n USnW U-n UR U S 9nUR UXS 9nU4U -n UR(aU W 4-n U $) Nra)rrrrrrrrerrrXrz'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`rr)rrhasattrrrrr)r8rrrrrrrreself_attn_past_key_valueself_attention_outputsrrpresent_key_valuecross_attn_present_key_valuecross_attn_past_key_valuecross_attention_outputsintermediate_output layer_outputs r;rxTFLayoutLMLayer.calls:H9S>"1#5Y] !%&)"&#'3/"0 " 2!4 ??,Qr2G 6r : ,QR0G'+$ ??4@4!122 =dVDDD @N?Yrs(;_c %&*&9&9--#&;'=8"3!': ' # 7q9  7" ==G,C2+F ( 14P P "//>N/O''-[TU]"S0UD6 Xl[UR5Vs/sHn[ USU3S9PM snUlgs snf)Nzlayer_._rr))r*r+r,rinum_hidden_layersrlayer)r8r,r9ir:s r;r+TFLayoutLMEncoder.__init__=sL "6" LQRXRjRjLklLkqofXaS>BLkl lsAc &U (aSOSn U(aSOSn U(aURR(aSOSnU(aSOSn[UR5H~unnU (aX4-n UbUUOSnU"UUUUUUUUU S9nUSnU(a UUS4- nU(dMJU US4-n URR(dMpUcMuUUS4-nM U (aX4-n U (d[ SXX455$[ UUU U US9$) Nr)rrrXrrac3.# UH ocMUv M g7fr{r)).0vs r; )TFLayoutLMEncoder.call..tscacs )last_hidden_statepast_key_valuesr attentionscross_attentions)r,r enumerater1tupler)r8rrrrrr; use_cacheroutput_hidden_states return_dictreall_hidden_statesall_attentionsall_cross_attentionsnext_decoder_cacher2 layer_moduler layer_outputss r;rxTFLayoutLMEncoder.callBsK#7BD0d%64;;;Z;Zr`d#,R$(4OA|#$58H$H!3B3N_Q/TXN(+-#A,&;'=-"3! M*!,M"}R'8&::"  !/=3C2E!E;;2227L7X+?=QRCSBU+U(1 56 14D D )nc ;+.+%1   r=cUR(agSUl[USS5bNURH=n[R"UR 5 UR S5 SSS5 M? gg!,(df  MR=f)NTr1)rOrPr1rJrKr&rQ)r8rSr1s r;rQTFLayoutLMEncoder.builds` ::  4$ ' 3]]5::.KK%/.$ 4..s A77 B )rOr,r1rzr)rrrrrrrr|rr|r;ztuple[tuple[tf.Tensor]] | Noner@ bool | Nonerr}rAr}rBr}rer}r~z>TFBaseModelOutputWithPastAndCrossAttentions | tuple[tf.Tensor]r{rrs@r;r.r.<sm"<  < "<  < 0 < !1 < 8< <  < #< < <  H< |&&r=r.c@^\rSrSrSU4SjjrSSjrSSjrSrU=r$) TFLayoutLMPooleric >[TU]"S0UD6 [RR UR [ UR5SSS9UlXl g)Ntanhr)rr activationr&r)) r*r+rr1rr-rr0rr,r7s r;r+TFLayoutLMPooler.__init__sR "6"\\''$$.v/G/GH (   r=c6USS2S4nURUS9nU$)Nrrb)r)r8rfirst_token_tensor pooled_outputs r;rxTFLayoutLMPooler.calls*+1a40 *< = r=c@UR(agSUl[USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  g=frrrRs r;rQTFLayoutLMPooler.buildr r )rOr,rrzr r{rrs@r;rNrNs HHr=rNc@^\rSrSrSU4SjjrSSjrSSjrSrU=r$) !TFLayoutLMPredictionHeadTransformic >[TU]"S0UD6 [RR UR [ UR5SS9Ul[UR[5(a[UR5Ul OURUl [RRURSS9UlXlg)Nrrr#r$r))r*r+rr1rr-rr0rrrrrtransform_act_fnr2r3r#r,r7s r;r+*TFLayoutLMPredictionHeadTransform.__init__s "6"\\''$$.v/G/GH(  f'' - -$5f6G6G$HD !$*$5$5D !88AVAV]h8i r=cdURUS9nURU5nURUS9nU$r)rr\r#rs r;rx&TFLayoutLMPredictionHeadTransform.calls8 - 8 --m< m< r=cHUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=frrrRs r;rQ'TFLayoutLMPredictionHeadTransform.buildrr)r#rOr,rr\rzr r{rrs@r;rZrZs" L Lr=rZch^\rSrSrS U4SjjrS SjrS SjrS SjrSSjrS Sjr SSjr S r U=r $)TFLayoutLMLMPredictionHeadic ~>[TU]"S0UD6 XlURUl[ USS9UlX lg)N transformrr))r*r+r,r-rZreinput_embeddingsr8r,rfr9r:s r;r+#TFLayoutLMLMPredictionHead.__init__s< "6" !--:6 T!1r=cjURURR4SSSS9UlUR(agSUl[ USS5bO[ R"URR5 URRS5 SSS5 gg!,(df  g=f)NzerosTbias)rArB trainabler&re) rLr,rMrkrOrPrJrKrer&rQrRs r;rQ TFLayoutLMLMPredictionHead.buildsOO4;;+A+A*CQXdhouOv ::  4d + 7t~~223$$T*43 833s >B$$ B2cUR$r{)rfr8s r;get_output_embeddings0TFLayoutLMLMPredictionHead.get_output_embeddingss$$$r=c^XRl[U5SURlgNr)rfr@rrMr8r[s r;set_output_embeddings0TFLayoutLMLMPredictionHead.set_output_embeddingss&',$+5e+r=rcc@^\rSrSrSU4SjjrSSjrSSjrSrU=r$) TFLayoutLMMLMHeadic D>[TU]"S0UD6 [XSS9Ulg)N predictionsrr))r*r+rcrrgs r;r+TFLayoutLMMLMHead.__init__s# "6"5fUbcr=c$URUS9nU$)Nr )r)r8sequence_outputprediction_scoress r;rxTFLayoutLMMLMHead.calls ,,?,K  r=cUR(agSUl[USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  g=f)NTr)rOrPrJrKrr&rQrRs r;rQTFLayoutLMMLMHead.build sb ::  4 - 9t//445  &&t,65 :55 A88 B)rOrr)rrr~rr{rrs@r;rrsd ! --r=rc^\rSrSr\rS S U4SjjjrS SjrS SjrSr \ S SSjj5r SSjr Sr U=r$)TFLayoutLMMainLayeric >[TU]"S0UD6 Xl[USS9Ul[ USS9UlU(a[USS9UlgSUlg)NrDrencoderpoolerr)) r*r+r,r rDr.rrNr)r8r,add_pooling_layerr9r:s r;r+TFLayoutLMMainLayer.__init__sL "6" .vLI(i@ AR&vH= X\ r=cUR$r{)rDros r;get_input_embeddings(TFLayoutLMMainLayer.get_input_embeddings!s r=c^XRl[U5SURlgrs)rDr@rrMrts r;set_input_embeddings(TFLayoutLMMainLayer.set_input_embeddings$s"!&%/%6q%9"r=c[e)z Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base class PreTrainedModel r)r8heads_to_prunes r; _prune_heads TFLayoutLMMainLayer._prune_heads(s "!r=cUbUb [S5eUb [U5nOUb[U5SSnO [S5eUc[R"USS9nUc[R"USS9nUc[R"US/-SS9nUR UUUUUU S9n[R "X>SSSUS45n[R "UURS 9n[R"S URS 9n[R"S URS 9n[R"[R"UU5U5nUb[eS/URR-nURUUUUSSS U U U U S 9 nUSnURbURUS9OSnU (d UU4USS-$[!UUUR"UR$UR&S9$)NzDYou cannot specify both input_ids and inputs_embeds at the same timerXz5You have to specify either input_ids or inputs_embedsrrYrr`)rkrlrmrnrorerg?gF) rrrrrr;r@rrArBrer )r: pooler_outputrr<r=)rrrJrgrDrrrconstantrsubtractrr,r0rrrrr<r=)r8rkrlrrnrmrrorrrrArBrerSembedding_outputextended_attention_maskone_cstten_thousand_cstencoder_outputsrrUs r;rxTFLayoutLMMainLayer.call/s"  ]%>cd d  "$Y/K  &$]3CR8KTU U  !WW+Q?N  !WW+Q?N <77 qc 1;D??%)' + #%**^!naQRT_`aTb=c"d#%''*AIYI_I_"`++c)9)?)?@;;x7G7M7MN"$++bkk'CZ.[]m"n  % %!>!>>I,,*2"7#' /!5#'  *!,FJkkF] / Bcg  #$ $ >-')77&11,==   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  N=f!,(df  N}=f!,(df  g=f)NTrDrr) rOrPrJrKrDr&rQrrrRs r;rQTFLayoutLMMainLayer.builds ::  4t , 8t334%%d+5 4D ) 5t||001 ""4(2 44 ( 4t{{//0 !!$'10 5 542110s$D0.E E0 D> E E )rOr,rDrr)T)r,rrr}rr NNNNNNNNNNNNF)rkTFModelInputType | Nonerlnp.ndarray | tf.Tensor | NonerrrnrrmrrrrorrrrrrrLrArLrBrLrer}r~ATFBaseModelOutputWithPoolingAndCrossAttentions | tuple[tf.Tensor]r{)rrrrr config_classr+rrrrrxrQrrrs@r;rrs!L]]:".2.28<8<6:377;?C@D)-,0#'c *c ,c 6 c 6 c 4 c 1c 5c  =c !>c 'c *c !c c  Kc c J ( (r=rc>^\rSrSrSr\rSr\U4Sj5r Sr U=r $)TFLayoutLMPreTrainedModelizz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. layoutlmcj>[TU]n[R"S[RSS9US'U$)N)NNr`rl)rArr&)r*input_signaturerJ TensorSpecint32)r8 signaturer:s r;r)TFLayoutLMPreTrainedModel.input_signatures/G+ MMrxxV\] &r=r)) rrrrrrrbase_model_prefixpropertyrrrrs@r;rrs' "L" r=raz This model inherits from [`TFPreTrainedModel`]. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a [keras.Model](https://www.tensorflow.org/api_docs/python/tf/keras/Model) subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior. TensorFlow models and layers in `transformers` accept two formats as input: - having all inputs as keyword arguments (like PyTorch models), or - having all inputs as a list, tuple or dict in the first positional argument. The reason the second format is supported is that Keras methods prefer this format when passing inputs to models and layers. Because of this support, when using methods like `model.fit()` things should "just work" for you - just pass your inputs and labels in any format that `model.fit()` supports! If, however, you want to use the second format outside of Keras methods like `fit()` and `predict()`, such as when creating your own layers or models with the Keras `Functional` API, there are three possibilities you can use to gather all the input Tensors in the first positional argument: - a single Tensor with `input_ids` only and nothing else: `model(input_ids)` - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: `model([input_ids, attention_mask])` or `model([input_ids, attention_mask, token_type_ids])` - a dictionary with one or several input Tensors associated to the input names given in the docstring: `model({"input_ids": input_ids, "token_type_ids": token_type_ids})` Note that when creating models and layers with [subclassing](https://keras.io/guides/making_new_layers_and_models_via_subclassing/) then you don't need to worry about any of this, as you can just pass inputs like you would to any other Python function! Args: config ([`LayoutLMConfig`]): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the [`~TFPreTrainedModel.from_pretrained`] method to load the model weights. a Args: input_ids (`Numpy array` or `tf.Tensor` of shape `({0})`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.__call__`] and [`PreTrainedTokenizer.encode`] for details. [What are input IDs?](../glossary#input-ids) bbox (`Numpy array` or `tf.Tensor` of shape `({0}, 4)`, *optional*): Bounding Boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings- 1]`. attention_mask (`Numpy array` or `tf.Tensor` of shape `({0})`, *optional*): Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. [What are attention masks?](../glossary#attention-mask) token_type_ids (`Numpy array` or `tf.Tensor` of shape `({0})`, *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 (`Numpy array` or `tf.Tensor` of shape `({0})`, *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) head_mask (`Numpy array` or `tf.Tensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*): Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. inputs_embeds (`tf.Tensor` of shape `({0}, 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. output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. training (`bool`, *optional*, defaults to `False`): Whether or not to use the model in training mode (some modules like dropout modules have different behaviors between training and evaluation). zbThe bare LayoutLM Model transformer outputting raw hidden-states without any specific head on top.c^\rSrSrSU4Sjjr\\"\RS55\ "\ \ S9S S Sjj555r S Sjr SrU=r$) TFLayoutLMModelicL>[TU]"U/UQ70UD6 [USS9Ulg)Nrr)r*r+rrr8r,rcr9r:s r;r+TFLayoutLMModel.__init__s( 3&3F3+FD r=batch_size, sequence_length output_typerc8URUUUUUUUU U U U S9 nU$)aX Returns: Examples: ```python >>> from transformers import AutoTokenizer, TFLayoutLMModel >>> import tensorflow as tf >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = TFLayoutLMModel.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="tf") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = tf.convert_to_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 ``` rkrlrrnrmrrorrArBre)r)r8rkrlrrnrmrrorrrrArBrers r;rxTFLayoutLMModel.call s@n--))%'/!5#  r=cUR(agSUl[USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  g=f)NTr)rOrPrJrKrr&rQrRs r;rQTFLayoutLMModel.buildgs^ ::  4T * 6t}}112 ##D)32 722r)rOrrzr)rkrrlrrrrnrrmrrrrorrrrrrrLrArLrBrLrerLr~rr{)rrrrr+rrLAYOUTLM_INPUTS_DOCSTRINGformatrr_CONFIG_FOR_DOCrxrQrrrs@r;rrs E *+D+K+KLi+jkBQ` .2.28<8<6:377;?C@D)-,0#' %@*@,@6 @ 6 @ 4 @1@5@ =@!>@'@*@!@@ K@l @D**r=rz6LayoutLM Model with a `language modeling` head on top.c^\rSrSr/SQrS U4SjjrS SjrS Sjr\\ "\ RS55\ "\ \S9SSSjj555rSS jrS rU=r$)TFLayoutLMForMaskedLMip)rcls.seq_relationshipzcls.predictions.decoder.weight nsp___clsc>[TU]"U/UQ70UD6 UR(a[R S5 [ USSS9Ul[XR RSS9Ul g)NzqIf you want to use `TFLayoutLMForMaskedLM` make sure `config.is_decoder=False` for bi-directional self-attention.Trrr& mlm___cls)rfr&) r*r+rloggerwarningrrrrDmlmrs r;r+TFLayoutLMForMaskedLM.__init__zs` 3&3F3    NN1  ,FdQ[\ $Vmm>V>V]hir=c.URR$r{)rrros r; get_lm_head!TFLayoutLMForMaskedLM.get_lm_headsxx###r=c[R"S[5 URS-URR-S-URR R-$)NzMThe method get_prefix_bias_name is deprecated. Please use `get_bias` instead./)warningswarn FutureWarningr&rrros r;get_prefix_bias_name*TFLayoutLMForMaskedLM.get_prefix_bias_namesG egtuyy3.4txx7K7K7P7PPPr=rrc URUUUUUUUUU U U S9 n U SnURXS9nU cSOURXS9nU (dU4U SS-nUbU4U-$U$[UUU RU R S9$)aF labels (`tf.Tensor` or `np.ndarray` 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]` Returns: Examples: ```python >>> from transformers import AutoTokenizer, TFLayoutLMForMaskedLM >>> import tensorflow as tf >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = TFLayoutLMForMaskedLM.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="tf") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = tf.convert_to_tensor([token_boxes]) >>> labels = tokenizer("Hello world", return_tensors="tf")["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 ```rr)rreNlabelsrralossrrr<)rrhf_compute_lossrrr<)r8rkrlrrnrmrrorrArBrrerrrrrs r;rxTFLayoutLMForMaskedLM.calls~--))%'/!5#  "!* HH_HX~t4+?+?v+?+h')GABK7F)-)9TGf$ Ev E$!//))   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTrr)rOrPrJrKrr&rQrrRs r;rQTFLayoutLMForMaskedLM.builds ::  4T * 6t}}112 ##D)3 4 % 1txx}}-t$.- 232.-r)rOrrrzr)r~r NNNNNNNNNNNF)rkrrlrrrrnrrmrrrrorrrLrArLrBrLrrrerLr~z#TFMaskedLMOutput | tuple[tf.Tensor]r{)rrrr"_keys_to_ignore_on_load_unexpectedr+rrrrrrrrrrxrQrrrs@r;rrps*& j$Q*+D+K+KLi+jk+;/Z.2.28<8<6:377;)-,0#'04 %V *V ,V 6 V 6 V 4 V 1V 5V 'V *V !V .V V  -V [lV p % %r=rz LayoutLM Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks. c^\rSrSr/SQrS/rS U4Sjjr\\"\ RS55\ "\ \ S9S S Sjj555rS SjrS rU=r$)#TFLayoutLMForSequenceClassificationi)rrcls.predictionsrr6cT>[TU]"U/UQ70UD6 URUl[USS9Ul[ R RURS9Ul [ R RUR[UR5SS9Ul Xlg)Nrrr' classifierrr*r+ num_labelsrrrr1r4r5r6rrr0rr,rs r;r+,TFLayoutLMForSequenceClassification.__init__s 3&3F3 +++FD ||++1K1K+L ,,,,##.v/G/GH-   r=rrc URUUUUUUUUU U U S9 n U SnURXS9nURUS9nU cSOURXS9nU (dU4U SS-nUbU4U-$U$[ UUU R U R S9$) aV labels (`tf.Tensor` or `np.ndarray` of shape `(batch_size,)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). Returns: Examples: ```python >>> from transformers import AutoTokenizer, TFLayoutLMForSequenceClassification >>> import tensorflow as tf >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = TFLayoutLMForSequenceClassification.from_pretrained("microsoft/layoutlm-base-uncased") >>> words = ["Hello", "world"] >>> normalized_word_boxes = [637, 773, 693, 782], [698, 773, 733, 782] >>> token_boxes = [] >>> for word, box in zip(words, normalized_word_boxes): ... word_tokens = tokenizer.tokenize(word) ... token_boxes.extend([box] * len(word_tokens)) >>> # add bounding boxes of cls + sep tokens >>> token_boxes = [[0, 0, 0, 0]] + token_boxes + [[1000, 1000, 1000, 1000]] >>> encoding = tokenizer(" ".join(words), return_tensors="tf") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = tf.convert_to_tensor([token_boxes]) >>> sequence_label = tf.convert_to_tensor([1]) >>> outputs = model( ... input_ids=input_ids, ... bbox=bbox, ... attention_mask=attention_mask, ... token_type_ids=token_type_ids, ... labels=sequence_label, ... ) >>> loss = outputs.loss >>> logits = outputs.logits ```rrrdrbNrrar)rr6rrr rr<)r8rkrlrrnrmrrorrArBrrerrUrrrs r;rx(TFLayoutLMForSequenceClassification.calls~--))%'/!5#     M M  6~t4+?+?v+?+]Y,F)-)9TGf$ Ev E)!//))   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=fNTrr rOrPrJrKrr&rQrr,r-rRs r;rQ)TFLayoutLMForSequenceClassification.buildj ::  4T * 6t}}112 ##D)3 4t , 8t334%%tT4;;3J3J&KL54 93254C+.3C<+ C9< D rOrr,r6rrrzr)rkrrlrrrrnrrmrrrrorrrLrArLrBrLrrrerLr~z-TFSequenceClassifierOutput | tuple[tf.Tensor]r{)rrrrr_keys_to_ignore_on_load_missingr+rrrrrr rrxrQrrrs@r;rrs*s&'1l# *+D+K+KLi+jk+ETcd.2.28<8<6:377;)-,0#'04 %W *W ,W 6 W 6 W 4 W 1W 5W 'W *W !W .W W  7W elW r M Mr=rz LayoutLM Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. c^\rSrSr/SQrS/rS U4Sjjr\\"\ RS55\ "\ \ S9S S Sjj555rS SjrS rU=r$) TFLayoutLMForTokenClassificationivrrrrrr6cV>[TU]"U/UQ70UD6 URUl[USSS9Ul[ R RURS9Ul [ R RUR[UR5SS9Ul Xlg)NTrrr'rrrrs r;r+)TFLayoutLMForTokenClassification.__init__s 3&3F3 +++FdQ[\ ||++1K1K+L ,,,,##.v/G/GH-   r=rrc URUUUUUUUUU U U S9 n U SnURXS9nURUS9nU cSOURXS9nU (dU4U SS-nUbU4U-$U$[ UUU R U R S9$) a labels (`tf.Tensor` or `np.ndarray` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`. Returns: Examples: ```python >>> import tensorflow as tf >>> from transformers import AutoTokenizer, TFLayoutLMForTokenClassification >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/layoutlm-base-uncased") >>> model = TFLayoutLMForTokenClassification.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="tf") >>> input_ids = encoding["input_ids"] >>> attention_mask = encoding["attention_mask"] >>> token_type_ids = encoding["token_type_ids"] >>> bbox = tf.convert_to_tensor([token_boxes]) >>> token_labels = tf.convert_to_tensor([1, 1, 0, 0]) >>> 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 ```rrrdrbNrrar)rr6rrr rr<)r8rkrlrrnrmrrorrArBrrerrrrrs r;rx%TFLayoutLMForTokenClassification.callsz--))%'/!5#  "!*,,o,Q8~t4+?+?v+?+]Y,F)-)9TGf$ Ev E&!//))   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=frrrRs r;rQ&TFLayoutLMForTokenClassification.buildrrrrzr)rkrrlrrrrnrrmrrrrorrrLrArLrBrLrrrerLr~z*TFTokenClassifierOutput | tuple[tf.Tensor]r{)rrrrrrr+rrrrrr rrxrQrrrs@r;rrvs*&(2l# *+D+K+KLi+jk+BQ`a.2.28<8<6:377;)-,0#'04 %U *U ,U 6 U 6 U 4 U 1U 5U 'U *U !U .U U  4U blU n M Mr=ra LayoutLM Model with a span classification head on top for extractive question-answering tasks such as [DocVQA](https://rrc.cvc.uab.es/?ch=17) (a linear layer on top of the final hidden-states output to compute `span start logits` and `span end logits`). c^\rSrSr/SQrS U4Sjjr\\"\RS55\ "\ \ S9S S Sjj555r S SjrSrU=r$) TFLayoutLMForQuestionAnsweringirc>[TU]"U/UQ70UD6 URUl[USSS9Ul[ R RUR[UR5SS9Ul Xl g)NTrr qa_outputsr) r*r+rrrrr1rrr0rr,rs r;r+'TFLayoutLMForQuestionAnswering.__init__st 3&3F3 +++FdQ[\ ,,,,##.v/G/GH-   r=rrcURUUUUUUUUU U U S9 nUSnURUS9n[R"USSS9unn[R"USS9n[R"USS9nSnU bU bS U 0nU US 'UR UUU4S 9nU (dUU4USS-nUbU4U-$U$[ UUUURURS 9$) aB start_positions (`tf.Tensor` or `np.ndarray` of shape `(batch_size,)`, *optional*): Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. end_positions (`tf.Tensor` or `np.ndarray` of shape `(batch_size,)`, *optional*): Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. Returns: Examples: ```python >>> import tensorflow as tf >>> from transformers import AutoTokenizer, TFLayoutLMForQuestionAnswering >>> from datasets import load_dataset >>> tokenizer = AutoTokenizer.from_pretrained("impira/layoutlm-document-qa", add_prefix_space=True) >>> model = TFLayoutLMForQuestionAnswering.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="tf" ... ) >>> 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"] = tf.convert_to_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[tf.math.argmax(start_scores, -1)[0]], word_ids[tf.math.argmax(end_scores, -1)[0]] >>> print(" ".join(words[start : end + 1])) M. Hamann P. Harper, P. Martinez ```rrrbrarX)r[num_or_size_splitsr_)inputr_Nstart_position end_positionr)r start_logits end_logitsrr<) rrrJsplitsqueezerr rr<)r8rkrlrrnrmrrorrArBstart_positions end_positionsrerrrrrrrrs r;rx#TFLayoutLMForQuestionAnswering.calls!L--))%'/!5#  "!*8#%88&QUW#X jzz 2> ZZjr:   &=+D&8F%2F> "''v|Z>X'YD"J/'!"+=F)-)9TGf$ Ev E-%!!//))   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=f)NTrr) rOrPrJrKrr&rQrr,r-rRs r;rQ$TFLayoutLMForQuestionAnswering.buildrr)rOr,rrrrzr)rkrrlrrrrnrrmrrrrorrrLrArLrBrLrrrrrerLr~z1TFQuestionAnsweringModelOutput | tuple[tf.Tensor]r{)rrrrrr+rrrrrr rrxrQrrrs@r;r r s!*& *+D+K+KLi+jk+IXgh.2.28<8<6:377;)-,0#'9=7; %h *h ,h 6 h 6 h 4 h 1h 5h 'h *h !h 7h 5h h  ;h ilh T M Mr=r )rrrr rrr)Dr __future__rrrnumpynp tensorflowrJactivations_tfrmodeling_tf_outputsrrrr r r modeling_tf_utilsr r rrrrrrrrtf_utilsrrrutilsrrrrconfiguration_layoutlmr get_loggerrrrr1Layerr rrrrrrr.rNrZrcrrrLAYOUTLM_START_DOCSTRINGrrrrrr __all__r)r=r;r-s" /   SRtt2   H %"C 5<<--C NAHell00AHJL5<<--L>0.%,,,,0.hHU\\//H<Lu||))L>d0ell((d0PK& **K&^Hu||))H<"L (:(:"LL-!3!3-b- **-(L(%,,,,L(L(^  1  (T5phS*/S* S*lRTlm@%57S@%n@%F xM*CEaxMxMv |M'@B[|M|M~  LM%>@WLMLM^ r=