cCiG>SrSSKJr SSKrSSKJr SSKrSSKr SSK J r SSK J r JrJrJrJrJrJrJr SSKJrJrJrJrJrJrJrJrJrJrJ r J!r! SS K"J#r#J$r$J%r% SS K&J'r'J(r(J)r)J*r*J+r+J,r, S S K-J.r. \+R^"\05r1S r2Sr3Sr4Sr5Sr6Sr7Sr8Sr9Sr:Sr;Sr"SS5r?"SS\RR5rB"SS\RR5rD"S S!\RR5rE\D\ES".rF"S#S$\RR5rG"S%S&\RR5rH"S'S(\RR5rI"S)S*\RR5rJ"S+S,\RR5rK"S-S.\RR5rL"S/S0\RR5rM"S1S2\RR5rN"S3S4\RR5rO"S5S6\RR5rP"S7S8\RR5rQ"S9S:\RR5rR"S;S<\RR5rS"S=S>\RR5rT"S?S@\RR5rU"SASB\RR5rV\ "SCSD\RR55rW"SESF\5rX\"SGSH\'55rYSIrZSJr[\)"SK\Z5"SLSM\X55r\\)"SN\Z5"SOSP\X\?55r]\)"SQ\Z5"SRSS\X\55r^"STSU\RR5r_\)"SV\Z5"SWSX\X\55r`\)"SY\Z5"SZS[\X\55ra\)"S\\Z5"S]S^\X\55rb\)"S_\Z5"S`Sa\X\55rc\)"Sb\Z5"ScSd\X\55rd/SeQreg)fzTF 2.0 MobileBERT model.) annotationsN) dataclass)get_tf_activation)TFBaseModelOutputTFBaseModelOutputWithPoolingTFMaskedLMOutputTFMultipleChoiceModelOutputTFNextSentencePredictorOutputTFQuestionAnsweringModelOutputTFSequenceClassifierOutputTFTokenClassifierOutput) TFMaskedLanguageModelingLossTFModelInputTypeTFMultipleChoiceLossTFNextSentencePredictionLossTFPreTrainedModelTFQuestionAnsweringLossTFSequenceClassificationLossTFTokenClassificationLossget_initializerkeraskeras_serializable unpack_inputs)check_embeddings_within_bounds shape_liststable_softmax) ModelOutputadd_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardloggingreplace_return_docstrings)MobileBertConfigzgoogle/mobilebert-uncasedr%z"vumichien/mobilebert-finetuned-nerzK['I-ORG', 'I-ORG', 'O', 'O', 'O', 'O', 'O', 'I-LOC', 'O', 'I-LOC', 'I-LOC']gQ?z%vumichien/mobilebert-uncased-squad-v2z'a nice puppet'gףp= @ zvumichien/emo-mobilebertz'others'z4.72c"\rSrSrSrSSjrSrg)TFMobileBertPreTrainingLossWz Loss function suitable for BERT-like pretraining, that is, the task of pretraining a language model by combining NSP + MLM. .. note:: Any label of -100 will be ignored (along with the corresponding logits) in the loss computation. c[RRS[RRRS9nU"[ R RUS5USS9n[ R"USS:gURS9nXE-n[ R"U5[ R"U5- nU"[ R RUS5US S9n[ R"USS:gURS9n X-n [ R"U 5[ R"U 5- n [ R"X{-S 5$) NT) from_logits reductionlabelsr)y_truey_predidtypenext_sentence_labelr$)r$) rlossesSparseCategoricalCrossentropy ReductionNONEtfnnrelucastr2 reduce_sumreshape) selfr.logitsloss_fnunmasked_lm_losses lm_loss_maskmasked_lm_lossesreduced_masked_lm_lossunmasked_ns_loss ns_loss_maskmasked_ns_lossreduced_masked_ns_losss o/home/james-whalen/.local/lib/python3.13/site-packages/transformers/models/mobilebert/modeling_tf_mobilebert.pyhf_compute_loss+TFMobileBertPreTrainingLoss.hf_compute_loss^s&,,<<Y^YeYeYoYoYtYt<u%BEEJJvh7G,HQWXYQZ[wwvh/47?Q?W?WX -<!#/?!@2==Q]C^!^#"%%**V|A\!\zz0I4PPN)r. tf.Tensorr?rNreturnrN)__name__ __module__ __qualname____firstlineno____doc__rJ__static_attributes__rMrLrIr)r)Ws  QrLr)c8^\rSrSrU4SjrSrSSjrSrU=r$)TFMobileBertIntermediatesc ">[TU]"S0UD6 [RR UR SS9Ul[UR[5(a[UR5Ul OURUl Xl g)NdensenamerM) super__init__rlayersDenseintermediate_sizerZ isinstance hidden_actstrrintermediate_act_fnconfigr>rfkwargs __class__s rIr^!TFMobileBertIntermediate.__init__tsm "6"\\''(@(@w'O f'' - -'89J9J'KD $'-'8'8D $ rLcJURU5nURU5nU$N)rZrer> hidden_statess rIcallTFMobileBertIntermediate.calls& =1 00? rLc@UR(agSUl[USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  g=fNTrZ) builtgetattrr8 name_scoperZr\buildrftrue_hidden_sizer> input_shapes rIrvTFMobileBertIntermediate.buildsm ::  4$ ' 3tzz/   $dkk.J.J!KL0/ 4// 3B B)rsrfrZrerl rPrQrRrSr^rorvrU __classcell__ris@rIrWrWss  MMrLrWc8^\rSrSrU4SjrSU4SjjrSrU=r$) TFLayerNormc2>Xl[TU]"U0UD6 grl) feat_sizer]r^)r>rargsrhris rIr^TFLayerNorm.__init__s" $)&)rLc>>[TU]SSUR/5 grl)r]rvrr>ryris rIrvTFLayerNorm.builds  tT4>>23rL)rrl)rPrQrRrSr^rvrUr}r~s@rIrrs*44rLrcB^\rSrSrSU4SjjrU4SjrSSjrSrU=r$)TFNoNormc 2>[TU]"S0UD6 Xlg)NrM)r]r^r)r>repsilonrhris rIr^TFNoNorm.__init__s "6""rLc>URSUR/SS9UlURSUR/SS9Ul[TU]U5 g)Nbiaszeros)shape initializerweightones) add_weightrrrr]rvrs rIrvTFNoNorm.buildsKOOF4>>2BPWOX ooht~~6FTZo[   k"rLc8XR-UR-$rl)rr)r>inputss rIro TFNoNorm.calls #dii//rL)rrrrl)rrN) rPrQrRrSr^rvrorUr}r~s@rIrrs## 00rLr) layer_normno_normc@^\rSrSrSrU4SjrSSjrSSjrSrU=r $) TFMobileBertEmbeddingszGConstruct the embeddings from word, position and token_type embeddings.c @>[TU]"S0UD6 URUlURUlXlUR UlUR UlURUl[RRUR SS9Ul [UR"UR URSS9Ul[RR!UR"S9UlURUR(aSOS-Ulg) Nembedding_transformationr[ LayerNormrr\)raterr$rM)r]r^ trigram_inputembedding_sizerf hidden_sizemax_position_embeddingsinitializer_rangerr_r`rNORM2FNnormalization_typelayer_norm_epsrDropouthidden_dropout_probdropoutembedded_input_sizergs rIr^TFMobileBertEmbeddings.__init__s "6"#11$33 !--'-'E'E$!'!9!9(- (:(:6;M;MTn(:(o%!!:!:;   (=(=K ||++1K1K+L #'#6#6t?Q?Q!WX#Y rLc [R"S5 URSURRUR /[ URS9S9UlSSS5 [R"S5 URSURRUR/[ URS9S9Ul SSS5 [R"S5 URSURUR/[ URS9S9Ul SSS5 UR(agSUl[US S5b[[R"UR R"5 UR R%SSUR&/5 SSS5 [US S5bO[R"UR(R"5 UR(R%S5 SSS5 gg!,(df  GN=f!,(df  GN]=f!,(df  GN=f!,(df  N=f!,(df  g=f) Nword_embeddingsr)r)r\rrtoken_type_embeddings embeddingsposition_embeddingsTrr)r8rurrf vocab_sizerrrrtype_vocab_sizerrrrrsrtrr\rvrrrxs rIrvTFMobileBertEmbeddings.builds ]], -//{{--t/B/BC+d>T>TU*DK.]]2 3)-!{{22D4D4DE+d>T>TU*9*D &4]]0 1'+!33T5E5EF+d>T>TU(7(D $2 ::  43T : Ft<<AAB--33T4AYAY4Z[C 4d + 7t~~223$$T*43 87. -4 32 1CB43s=AH =AH#>H-:)H?#I H H*- H<? I  Ic HUcUceUb>[XRR5 [R"UR US9n[ U5SSnUc[R"USS9nUR(aT[R"[R"USS2SS24S5U[R"USS2SS24S5/S S 9nUR(dURUR:waURU5nUc+[R"[R"SUSS 9SS 9n[R"UR US9n[R"UR"US9nXG-U-n UR%U S 9n UR'XS 9n U $)zr Applies embedding based on inputs tensor. Returns: final_embeddings (`tf.Tensor`): output embedding tensor. N)paramsindicesr)dimsvaluer$)rr)rr$r)r)r$rraxis)startlimit)r)rtraining)rrfrr8gatherrrfillrconcatpadrrr expand_dimsrangerrrr) r> input_ids position_idstoken_type_ids inputs_embedsrryposition_embedstoken_type_embedsfinal_embeddingss rIroTFMobileBertEmbeddings.calls%-*?@@  *9kk6L6L MIIT[[)LM /4  !WW+Q?N   IIFF=AB/1IJ!FF=CRC02JK  M   !4!48H8H!H 99-HM  >>"((+b/*RYZ))4+C+C\ZIIT-G-GQ_`(:=NN>>1A>B<[TU]"S0UD6 URUR-S:wa%[ SURSUR35eURUlUR UlURUR-S:Xde[ URUR- 5UlURUR-Ul [RRUR[UR5SS9Ul[RRUR[UR5SS9Ul[RRUR[UR5SS9Ul[RR%UR&5UlXlg) NrzThe hidden size (z6) is not a multiple of the number of attention heads (querykernel_initializerr\keyrrM)r]r^rnum_attention_heads ValueErroroutput_attentionsintrwattention_head_size all_head_sizerr_r`rrrrrrattention_probs_dropout_probrrfrgs rIr^"TFMobileBertSelfAttention.__init__s "6"    : : :a ?#F$6$6#78 4457  $*#=#= !'!9!9!!F$>$>>!CCC#&v'>'>A[A['[#\ !558P8PP\\''   ?6C[C[3\cj(  <<%%   ?6C[C[3\ch& \\''   ?6C[C[3\cj(  ||++F,O,OP  rLc[R"XSURUR45n[R"U/SQS9$)Nrrrr$rperm)r8r=rr transpose)r>x batch_sizes rItranspose_for_scores.TFMobileBertSelfAttention.transpose_for_scores,s6 JJqr4+C+CTE]E]^ _||AL11rLc[U5SnURU5n URU5n URU5n UR X5n UR X5n UR X5n[ R "XSS9n[ R"[U 5SURS9nU[ RRU5- nUb"[ R"XORS9nX-n[USS9nURUUS9nUbUU-n[ R "UU5n[ R"U/SQS 9n[ R"UUSUR45nU(aUU4nU$U4nU$) NrT) transpose_brr1rrrr)rrrrrr8matmulr;r2mathsqrtrrrr=r)r> query_tensor key_tensor value_tensorattention_mask head_maskrrrmixed_query_layermixed_key_layermixed_value_layer query_layer key_layer value_layerattention_scoresdkattention_probs context_layeroutputss rIroTFMobileBertSelfAttention.call1s /2  JJ|4((:. JJ|4//0AN --oJ //0AN 99  WWZ *2.6F6L6L M+bggll2.>>  %WW^;Q;QRN/@ ))9C,,,J  - 9O /;?  ]F  JD,>,>? 7H=/2O\M]rLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5b[R"URR 5 URR SSURR(aURROURR/5 SSS5 gg!,(df  GN(=f!,(df  N=f!,(df  g=f)NTrrr) rsrtr8rurr\rvrfrwrruse_bottleneck_attentionrrxs rIrvTFMobileBertSelfAttention.build^s8 ::  4$ ' 3tzz/   $dkk.J.J!KL0 4 % 1txx}}-dDKK,H,HIJ. 4$ ' 3tzz/   ;;?? 44![[44 0/ 4 0/.-0/s%3F&3F88A$G & F58 G G) rrrsrfrrrrrrFrl) rPrQrRrSr^rrorvrUr}r~s@rIrrs$62 ns+ZrLrc<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFMobileBertSelfOutputiuc >[TU]"S0UD6 URUl[RR UR [UR5SS9Ul [UR"UR URSS9Ul UR(d.[RRUR5UlXlg)NrZrrrrM)r]r^use_bottleneckrr_r`rwrrrZrrrrrrrrfrgs rIr^TFMobileBertSelfOutput.__init__vs "6"$33\\''  # #H`H`8aho(  !!:!:;  # #V-B-B "" <<//0J0JKDL rLcURU5nUR(dURXS9nURX-5nU$Nr)rZr rr)r>rnresidual_tensorrs rIroTFMobileBertSelfOutput.calls@ =1 "" LLLJM}'FG rLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=fNTrZr rsrtr8rurZr\rvrfrwrrxs rIrvTFMobileBertSelfOutput.build ::  4$ ' 3tzz/   $dkk.J.J!KL0 4d + 7t~~223$$T*43 80/433C+C<+ C9< D )rrsrfrZrr rrlr|r~s@rIr r us  + +rLr cD^\rSrSrU4SjrSrSSjrSSjrSrU=r $) TFMobileBertAttentionic b>[TU]"S0UD6 [USS9Ul[ USS9Ulg)Nr>r[outputrM)r]r^rr>r mobilebert_outputrgs rIr^TFMobileBertAttention.__init__s0 "6"-f6B !7X!NrLc[erlNotImplementedError)r>headss rI prune_heads!TFMobileBertAttention.prune_headss!!rLc bURXX5XgUS9n URU SXHS9n U 4U SS-n U $)Nrrr$)r>r) r>rrr layer_inputrrrr self_outputsattention_outputrs rIroTFMobileBertAttention.callsVyy lIks!  11,q/;1b#% QR(88rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTr>r)rsrtr8rur>r\rvrrxs rIrvTFMobileBertAttention.builds ::  4 & 2tyy~~. %/ 4,d 3 ?t55::;&&,,T2<; @/.<;C.C% C"% C3)rsrr>rrl) rPrQrRrSr^r!rorvrUr}r~s@rIrrs"O "& 3 3rLrc<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFOutputBottleneckic L>[TU]"S0UD6 [RR UR SS9Ul[UR"UR URSS9Ul [RRUR5Ul XlgNrZr[rrrM)r]r^rr_r`rrZrrrrrrrrfrgs rIr^TFOutputBottleneck.__init__s "6"\\''(:(:'I  !:!:;   (=(=K ||++F,F,FG  rLclURU5nURXCS9nURXB-5nU$r)rZrr)r>rnrr layer_outputss rIroTFOutputBottleneck.calls7 =1  ] F }'FG rLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=frrrxs rIrvTFOutputBottleneck.buildrr)rrsrfrZrrrlr|r~s@rIr,r,s + +rLr,c<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFMobileBertOutputic >[TU]"S0UD6 URUl[RR UR [UR5SS9Ul [UR"UR URSS9Ul UR(d/[RRUR5UlO[#USS9UlXlg)NrZrrr bottleneckr[rM)r]r^r rr_r`rwrrrZrrrrrrrr,r8rfrgs rIr^TFMobileBertOutput.__init__s "6"$33\\''  # #H`H`8aho(  !!:!:;  # #V-B-B "" <<//0J0JKDL0lKDO rLcURU5nUR(d$URXS9nURX-5nU$URX-5nUR X5nU$r)rZr rrr8)r>rnresidual_tensor_1residual_tensor_2rs rIroTFMobileBertOutput.callsi =1 "" LLLJM NN=+LMM!NN=+LMM OOMMMrLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 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)NTrZrr8) rsrtr8rurZr\rvrfrarr8rxs rIrvTFMobileBertOutput.builds ::  4$ ' 3tzz/   $dkk.K.K!LM0 4d + 7t~~223$$T*4 4t , 8t334%%d+54 9 0/4354s$3EE!E) E E&) E7)rr8rsrfrZrr rrlr|r~s@rIr6r6s  , ,rLr6c8^\rSrSrU4SjrSrSSjrSrU=r$)TFBottleneckLayeric >[TU]"S0UD6 [RR UR SS9Ul[UR"UR URSS9Ul Xl gr.) r]r^rr_r`intra_bottleneck_sizerZrrrrrfrgs rIr^TFBottleneckLayer.__init__se "6"\\''(D(D7'S  !:!:;  ( (&2G2Gk  rLcJURU5nURU5nU$rlrZr)r>rrns rIroTFBottleneckLayer.call s$ 6* }5 rLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=fr rsrtr8rurZr\rvrfrrrxs rIrvTFBottleneckLayer.build ::  4$ ' 3tzz/   $dkk.E.E!FG0 4d + 7t~~223$$T*43 80/43rrrsrfrZrlr|r~s@rIrArA + +rLrAc8^\rSrSrU4SjrSrSSjrSrU=r$) TFBottleneckic >[TU]"S0UD6 URUlURUl[ USS9UlUR(a[ USS9Ulgg)Ninputr[ attentionrM)r]r^key_query_shared_bottleneckrrAbottleneck_inputrRrgs rIr^TFBottleneck.__init__sZ "6"+1+M+M((.(G(G% 1&w G  + +.vKHDN ,rLcURU5nUR(aU4S-$UR(aURU5nX3X4$XX4$)N)rTrrSrR)r>rnbottlenecked_hidden_statesshared_attention_inputs rIroTFBottleneck.call&sZ"&*%:%:=%I"  ( (.014 4  - -%)^^M%B "*Mn n!-\ \rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTrTrR)rsrtr8rurTr\rvrRrxs rIrvTFBottleneck.build@s ::  4+T 2 >t4499:%%++D1; 4d + 7t~~223$$T*43 8;:43r*)rRrTrsrSrrlr|r~s@rIrOrOsI]4 + +rLrOc8^\rSrSrU4SjrSrSSjrSrU=r$) TFFFNOutputiLc >[TU]"S0UD6 [RR UR SS9Ul[UR"UR URSS9Ul Xl gr.) r]r^rr_r`rwrZrrrrrfrgs rIr^TFFFNOutput.__init__Mse "6"\\''(?(?g'N  !:!:;  # #V-B-B  rLcNURU5nURX-5nU$rlrF)r>rnrs rIroTFFFNOutput.callUs' =1 }'FG rLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=fr) rsrtr8rurZr\rvrfrarrxs rIrvTFFFNOutput.buildZs ::  4$ ' 3tzz/   $dkk.K.K!LM0 4d + 7t~~223$$T*43 80/43rrLrlr|r~s@rIr^r^LrMrLr^c8^\rSrSrU4SjrSrSSjrSrU=r$) TFFFNLayerifc b>[TU]"S0UD6 [USS9Ul[ USS9Ulg)N intermediater[rrM)r]r^rWrhr^rrgs rIr^TFFFNLayer.__init__gs1 "6"4V.Q!,V(!CrLcJURU5nURX!5nU$rl)rhr)r>rnintermediate_outputr1s rIroTFFFNLayer.callls*"// >../BR rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTrhr)rsrtr8rurhr\rvrrxs rIrvTFFFNLayer.buildqs ::  4 . :t00556!!''-7 4,d 3 ?t55::;&&,,T2<; @76<;r*)rsrhrrlr|r~s@rIrfrffsD  3 3rLrfc<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFMobileBertLayeri}c >[TU]"S0UD6 URUlURUl[ USS9Ul[ USS9Ul[USS9Ul UR(a[USS9Ul URS:a:[URS- 5Vs/sHn[USU3S9PM snUlggs snf) NrRr[rhrr8r$zffn.rM)r]r^r num_feedforward_networksrrRrWrhr6rrOr8rrfffnr>rfrhiris rIr^TFMobileBertLayer.__init__~s "6"$33(.(G(G%.vKH4V.Q!3F!J   *6 EDO  * *Q .EJ6KjKjmnKnEopEo 6$qc ;EopDH /ps+C c UR(aURU5upgpO U/S-upgpURUUUU UUUUS9n U Sn U 4n URS:wa+[ UR 5HupU"U 5n X4- n M UR U 5nURXXS9nU4U SS-[R"S5UUUU U U4-U -nU$)NrWrrr$) r r8rRrr enumeratersrhrr8constant)r>rnrrrrrrrr$attention_outputsr&sru ffn_modulerk layer_outputrs rIroTFMobileBertLayer.calls.   BF//R_B` ?LlKCP/TUBU ?Ll NN       +  -Q/    ( (A -!*488!4 #-.>#? (("5#//0@A--.AUb-v O# $ A #   rLc~UR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bNURH=n[R"UR 5 UR S5 SSS5 M? gg!,(df  GN=f!,(df  GN6=f!,(df  N=f!,(df  N=f!,(df  M=f)NTrRrhrr8rs) rsrtr8rurRr\rvrhrr8rsr>rylayers rIrvTFMobileBertLayer.builds ::  4d + 7t~~223$$T*4 4 . :t00556!!''-7 4,d 3 ?t55::;&&,,T2< 4t , 8t334%%d+5 4 % 1]]5::.KK%/." 24376<;54/.s<G'.G9 H &HH-' G69 H H H*- H< )rRr8rsrsrhrrrr rrlr|r~s@rIrprp}s q+Z&&rLrpc>^\rSrSrU4SjrSSjrSSjrSrU=r$)TFMobileBertEncoderic >[TU]"S0UD6 URUlURUl[ UR 5Vs/sHn[ USU3S9PM snUlgs snf)Nzlayer_._r[rM)r]r^routput_hidden_statesrnum_hidden_layersrprrts rIr^TFMobileBertEncoder.__init__sd "6"!'!9!9$*$?$?!NSTZTlTlNmnNm'xs^DNmn ns A+c $U(aSOSnU(aSOSn [UR5H1upU(aX4-nU "XX:XGS9n U SnU(dM)XS4-n M3 U(aX4-nU(d[SXU 455$[XU S9$)NrMrrr$c3.# UH ocMUv M g7frlrM).0vs rI +TFMobileBertEncoder.call..sh$Vq$Vs )last_hidden_statern attentions)rxrtupler) r>rnrrrr return_dictrall_hidden_statesall_attentionsru layer_moduler1s rIroTFMobileBertEncoder.calls#7BD0d(4OA#$58H$H!(y|=NM*!,M  !/3C2E!E 5 14D D h]~$Vhh h +Yg  rLcUR(agSUl[USS5bNURH=n[R"UR 5 UR S5 SSS5 M? gg!,(df  MR=f)NTr)rsrtrr8rur\rvrs rIrvTFMobileBertEncoder.builds` ::  4$ ' 3]]5::.KK%/.$ 4..s A77 B )rsrrrrrlr|r~s@rIrrso! F&&rLrc8^\rSrSrU4SjrSrSSjrSrU=r$)TFMobileBertPooleric >[TU]"S0UD6 URUlUR(aB[R R UR[UR5SSS9Ul Xl g)NtanhrZ)r activationr\rM) r]r^classifier_activation do_activaterr_r`rrrrZrfrgs rIr^TFMobileBertPooler.__init__si "6"!77   ++""#263K3K#L! ,DJ  rLc`USS2S4nUR(dU$URU5nU$Nr)rrZ)r>rnfirst_token_tensor pooled_outputs rIroTFMobileBertPooler.calls7+1a40% % JJ'9:M rLc@UR(agSUl[USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  g=frr) rsrtr8rurZr\rvrfrrxs rIrvTFMobileBertPooler.buildsm ::  4$ ' 3tzz/   $dkk.E.E!FG0/ 4//r{)rsrfrZrrlr|r~s@rIrrs !HHrLrc8^\rSrSrU4SjrSrSSjrSrU=r$)#TFMobileBertPredictionHeadTransformi"c >[TU]"S0UD6 [RR UR [ UR5SS9Ul[UR[5(a[UR5Ul OURUl [S"UR URSS9UlXlg)NrZrrrrrM)r]r^rr_r`rrrrZrbrcrdrtransform_act_fnrrrrfrgs rIr^,TFMobileBertPredictionHeadTransform.__init__#s "6"\\''   ?6C[C[3\cj(   f'' - -$5f6G6G$HD !$*$5$5D ! .v/A/A6K`K`grs rLclURU5nURU5nURU5nU$rl)rZrrrms rIro(TFMobileBertPredictionHeadTransform.call/s4 =1 --m< }5 rLcUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=frrIrxs rIrv)TFMobileBertPredictionHeadTransform.build5rKr)rrsrfrZrrlr|r~s@rIrr"s  + +rLrcP^\rSrSrU4SjrS SjrSrSrSrSr Sr S r U=r $) TFMobileBertLMPredictionHeadiAc P>[TU]"S0UD6 [USS9UlXlg)N transformr[rM)r]r^rrrfrgs rIr^%TFMobileBertLMPredictionHead.__init__Bs& "6"s rIget_output_embeddings2TFMobileBertLMPredictionHead.get_output_embeddings]s rLcJXl[U5SURlgr)rrrfrr>rs rIset_output_embeddings2TFMobileBertLMPredictionHead.set_output_embeddings`s !+E!21!5 rLcSUR0$)Nr)rrs rIget_bias%TFMobileBertLMPredictionHead.get_biasds ""rLcXUSUl[US5SURlg)Nrr)rrrfrrs rIset_bias%TFMobileBertLMPredictionHead.set_biasgs'&M !+E&M!:1!= rLcURU5n[R"U[R"[R"UR 5UR /SS95nXR-nU$)Nrr)rr8rrrrrZrrms rIro!TFMobileBertLMPredictionHead.callksW}5  -BLLrLrc8^\rSrSrU4SjrSrSSjrSrU=r$)TFMobileBertMLMHeadirc D>[TU]"S0UD6 [USS9Ulg)N predictionsr[rM)r]r^rrrgs rIr^TFMobileBertMLMHead.__init__ss" "6"7]SrLc(URU5nU$rlr)r>sequence_outputprediction_scoress rIroTFMobileBertMLMHead.callws ,,_=  rLcUR(agSUl[USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  g=f)NTr)rsrtr8rurr\rvrxs rIrvTFMobileBertMLMHead.build{sb ::  4 - 9t//445  &&t,65 :55 A88 B)rsrrlr|r~s@rIrrrsT!--rLrct^\rSrSr\rS U4SjjrSrSrSr \ S Sj5r S Sjr Sr U=r$) TFMobileBertMainLayeric 2>[TU]"S0UD6 XlURUlURUlUR UlUR Ul[USS9Ul [USS9Ul U(a[USS9Ul gSUl g)Nrr[encoderpoolerrM)r]r^rfrrruse_return_dictrrrrrrr)r>rfadd_pooling_layerrhris rIr^TFMobileBertMainLayer.__init__s "6" !'!9!9!'!9!9$*$?$?!!110lK*6 B CT(h? Z^ rLcUR$rl)rrs rIget_input_embeddings*TFMobileBertMainLayer.get_input_embeddingss rLc^XRl[U5SURlgr)rrrrrs rIset_input_embeddings*TFMobileBertMainLayer.set_input_embeddingss"!&%/%6q%9"rLc[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)r>heads_to_prunes rI _prune_heads"TFMobileBertMainLayer._prune_headss "!rLc \UbUb [S5eUb [U5n OUb[U5SSn O [S5eUc[R"U S5nUc[R"U S5nUR XX6U S9n [R "X+SSSU S45n [R "XRS9n [R"SU RS9n[R"S U RS9n[R"[R"X5U5n Ub[eS/UR-nURU U UUUU U S9nUSnURbURU5OSnU (d UU4USS-$[UUUR UR"S 9$) NzDYou cannot specify both input_ids and inputs_embeds at the same timerz5You have to specify either input_ids or inputs_embedsr$rrr1g?g)r pooler_outputrnr)rrr8rrr=r;r2rymultiplysubtractrrrrrrnr)r>rrrrrrrrrrryembedding_outputextended_attention_maskone_cstten_thousand_cstencoder_outputsrrs rIroTFMobileBertMainLayer.calls  ]%>cd d  "$Y/K  &$]3CR8KTU U  !WW[!4N  !WW[!4N??9Nlt?u#%**^!naQRT_`aTb=c"d#%''*AI_I_"`++c)9)?)?@;;x7G7M7MN"$++bkk'.[]m"n  % %!7!77I,,  #   ' *!,8< 8O O4UY  #$ $ ,-')77&11   rLcUR(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)NTrrr) rsrtr8rurr\rvrrrxs rIrvTFMobileBertMainLayer.builds ::  4t , 8t334%%d+5 4D ) 5t||001 ""4(2 44 ( 4t{{//0 !!$'10 5 542110$D0.E E0 D> E E ) rsrfrrrrrrr)T NNNNNNNNNFrl)rPrQrRrSr% config_classr^rrrrrorvrUr}r~s@rIrrs^#L _:"!Q Q f ( (rLrc \rSrSrSr\rSrSrg)TFMobileBertPreTrainedModelizz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. mobilebertrMN) rPrQrRrSrTr%rbase_model_prefixrUrMrLrIrrs $L$rLrc`\rSrSr%SrSrS\S'SrS\S'SrS\S'Sr S\S 'Sr S\S 'S r g) TFMobileBertForPreTrainingOutputia Output type of [`TFMobileBertForPreTraining`]. Args: prediction_logits (`tf.Tensor` of shape `(batch_size, sequence_length, config.vocab_size)`): Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). seq_relationship_logits (`tf.Tensor` of shape `(batch_size, 2)`): Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation before SoftMax). hidden_states (`tuple(tf.Tensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `tf.Tensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs. attentions (`tuple(tf.Tensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`): Tuple of `tf.Tensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length, sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads. Nztf.Tensor | Nonelossprediction_logitsseq_relationship_logitsztuple[tf.Tensor] | NonernrrM) rPrQrRrSrTr__annotations__rrrnrrUrMrLrIrrsB,"D !*.'.04-4-1M*1*.J'.rLra 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! Parameters: config ([`MobileBertConfig`]): 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 [`~PreTrainedModel.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) 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. This argument can be used only in eager mode, in graph mode the value in the config will be used instead. 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. This argument can be used only in eager mode, in graph mode the value in the config will be used instead. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. This argument can be used in eager mode, in graph mode the value will always be set to True. 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). zdThe bare MobileBert Model transformer outputting raw hidden-states without any specific head on top.c^\rSrSrU4Sjr\\"\RS55\ "\ \ \ S9SS Sjj555r S SjrSrU=r$) TFMobileBertModelicL>[TU]"U/UQ70UD6 [USS9Ulg)Nrr[)r]r^rrr>rfrrhris rIr^TFMobileBertModel.__init__s( 3&3F3/\JrLbatch_size, sequence_length checkpoint output_typerc 6URUUUUUUUUU U S9 n U $)N) rrrrrrrrrr)r) r>rrrrrrrrrrrs rIroTFMobileBertModel.calls<(//))%'/!5#"  rLcUR(agSUl[USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  g=f)NTr)rsrtr8rurr\rvrxs rIrvTFMobileBertModel.builds^ ::  4t , 8t334%%d+54 944r)rsrr)rTFModelInputType | Nonernp.ndarray | tf.Tensor | Nonerrrrrrrrr bool | NonerrrrrrrOz$tuple | TFBaseModelOutputWithPoolingrl)rPrQrRrSr^rr!MOBILEBERT_INPUTS_DOCSTRINGformatr_CHECKPOINT_FOR_DOCr_CONFIG_FOR_DOCrorvrUr}r~s@rIrrs K*+F+M+MNk+lm&0$.28<8<6:377;)-,0#' %*66  4  1 5'*! . n8,,rLrz MobileBert Model with two heads on top as done during the pretraining: a `masked language modeling` head and a `next sentence prediction (classification)` head. c^\rSrSrU4SjrSrSr\\"\ RS55\ "\ \ S9S S Sjj555rS SjrS rS rU=r$)TFMobileBertForPreTrainingic>[TU]"U/UQ70UD6 [USS9Ul[ USS9Ul[ USS9Ulg)Nrr[predictions___clsseq_relationship___cls)r]r^rrrrTFMobileBertOnlyNSPHeadseq_relationshiprs rIr^#TFMobileBertForPreTraining.__init__sH 3&3F3/\J.v>> import tensorflow as tf >>> from transformers import AutoTokenizer, TFMobileBertForPreTraining >>> tokenizer = AutoTokenizer.from_pretrained("google/mobilebert-uncased") >>> model = TFMobileBertForPreTraining.from_pretrained("google/mobilebert-uncased") >>> input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1 >>> outputs = model(input_ids) >>> prediction_scores, seq_relationship_scores = outputs[:2] ``` rrrrrrrrrNrr.r3r.r?)rrrrnr)rrrrJrrnr)r>rrrrrrrrrr.r3rrrrrseq_relationship_score total_lossd_labelsrs rIroTFMobileBertForPreTraining.callsB// ))%'/!5#"  *1!& ,,_=!%!6!6}!E  "5"A &)H.AH* +--XGXZpFq-rJ')?@712;NF/9/EZMF* Q6 Q//$:!//))   rLcUR(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)NTrrr) rsrtr8rurr\rvrrrxs rIrv TFMobileBertForPreTraining.builds ::  4t , 8t334%%d+5 4 - 9t//445  &&t,6 4+T 2 >t4499:%%++D1;: ? 5465;:rcUS:XaUS4$U4$Nzcls.predictions.decoder.weightz,mobilebert.embeddings.word_embeddings.weightrMr> tf_weights rItf_to_pt_weight_rename1TFMobileBertForPreTraining.tf_to_pt_weight_rename- 8 8LL L< rL)rsrrr NNNNNNNNNNNF)rrrrrrrrrrrrrrrrrrr.rr3rrrrOz(tuple | TFMobileBertForPreTrainingOutputrl)rPrQrRrSr^rr(rr!rrr#rrrorvr8rUr}r~s@rIrrs _ ,a*+F+M+MNk+lm+KZij.28<8<6:377;)-,0#'04=A %? *? 6? 6 ? 4 ? 1 ? 5? '? *? !? .? ;? ?  2? kn? B 2  rLrz8MobileBert Model with a `language modeling` head on top.c ^\rSrSr/SQrU4SjrSrSr\\ "\ RS55\ "\ \\SSS 9SSS jj555rSS jrS rS rU=r$)TFMobileBertForMaskedLMi4)rrcls.seq_relationshipcl>[TU]"U/UQ70UD6 [USSS9Ul[ USS9Ulg)NFrrr\rr[)r]r^rrrrrs rIr^ TFMobileBertForMaskedLM.__init__=s; 3&3F3/%Vbc.vrrrrrrrrrr.rrrrrrs rIroTFMobileBertForMaskedLM.callJs:// ))%'/!5#"  "!* ,,_,P~t4+?+?+Z')GABK7F)-)9TGf$ Ev E$!//))   rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTrr)rsrtr8rurr\rvrrxs rIrvTFMobileBertForMaskedLM.builds ::  4t , 8t334%%d+5 4 - 9t//445  &&t,65 :5465r*cUS:XaUS4$U4$r5rMr6s rIr8.TFMobileBertForMaskedLM.tf_to_pt_weight_renamer:rL)rsrr NNNNNNNNNNF)rrrrrrrrrrrrrrrrrrr.rrrrOztuple | TFMaskedLMOutputrl)rPrQrRrS"_keys_to_ignore_on_load_unexpectedr^rr(rr!rrrrr rrorvr8rUr}r~s@rIr=r=4s*& Q ,Q*+F+M+MNk+lm&$$! .28<8<6:377;)-,0#'04 %. *. 6. 6 . 4 . 1 . 5. '. *. !. .. .  ". n. ` -  rLr=c8^\rSrSrU4SjrSrSSjrSrU=r$)ric x>[TU]"S0UD6 [RR SSS9UlXlg)Nrrr[rM)r]r^rr_r`rrfrgs rIr^ TFMobileBertOnlyNSPHead.__init__s5 "6" % 2 21;M 2 N rLc(URU5nU$rl)r)r>rr.s rIroTFMobileBertOnlyNSPHead.calls!%!6!6}!E%%rLc@UR(agSUl[USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  g=f)NTr) rsrtr8rurr\rvrfrrxs rIrvTFMobileBertOnlyNSPHead.buildst ::  4+T 2 >t4499:%%++T49P9P,QR;: ?::r{)rsrfrrlr|r~s@rIrrs &SSrLrzPMobileBert Model with a `next sentence prediction (classification)` head on top.c^\rSrSrSS/rU4Sjr\\"\RS55\ "\ \ S9S S Sjj555r S SjrS rU=r$) %TFMobileBertForNextSentencePredictionircls.predictionscj>[TU]"U/UQ70UD6 [USS9Ul[ USS9Ulg)Nrr[r)r]r^rrrclsrs rIr^.TFMobileBertForNextSentencePrediction.__init__s7 3&3F3/\J*68PQrLrr*c URUUUUUUUUU U S9 n U Sn URU 5nU cSOURXS9nU (dU4U SS-nUbU4U-$U$[UUU RU R S9$)a Return: Examples: ```python >>> import tensorflow as tf >>> from transformers import AutoTokenizer, TFMobileBertForNextSentencePrediction >>> tokenizer = AutoTokenizer.from_pretrained("google/mobilebert-uncased") >>> model = TFMobileBertForNextSentencePrediction.from_pretrained("google/mobilebert-uncased") >>> prompt = "In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced." >>> next_sentence = "The sky is blue due to the shorter wavelength of blue light." >>> encoding = tokenizer(prompt, next_sentence, return_tensors="tf") >>> logits = model(encoding["input_ids"], token_type_ids=encoding["token_type_ids"])[0] ```r,r$Nr-rrK)rr^rJr rnr)r>rrrrrrrrrr3rrrseq_relationship_scoresnext_sentence_lossrs rIro*TFMobileBertForNextSentencePrediction.callsF// ))%'/!5#"    "&((="9#* %%-@%a  -/'!"+=F7I7U')F2 a[a a,#*!//))   rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  Nl=f!,(df  g=f)NTrr^)rsrtr8rurr\rvr^rxs rIrv+TFMobileBertForNextSentencePrediction.builds ::  4t , 8t334%%d+5 4 % 1txx}}-t$.- 254.-r*)rsr^rrQ)rrrrrrrrrrrrrrrrrrr3rrrrOz%tuple | TFNextSentencePredictorOutputrl)rPrQrRrSrRr^rr!rrr#r rrorvrUr}r~s@rIr[r[s +?@R)S&R *+F+M+MNk+lm+HWfg.28<8<6:377;)-,0#'=A %> *> 6> 6 > 4 > 1 > 5> '> *> !> ;> >  /> hn> @ % %rLr[z MobileBert 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/rU4Sjr\\"\ RS55\ "\ \ \\\S9S S Sjj555rS SjrS rU=r$) %TFMobileBertForSequenceClassificationirrr\r>rc>[TU]"U/UQ70UD6 URUl[USS9UlUR b UR O UR n[RRU5Ul [RRUR[UR5SS9UlXlg)Nrr[ classifierrr]r^ num_labelsrrclassifier_dropoutrrr_rrr`rrrjrfr>rfrrhrmris rIr^.TFMobileBertForSequenceClassification.__init__s 3&3F3 ++/\J)/)B)B)NF % %TZTnTn ||++,>? ,,,,   /&BZBZ2[bn-  rLrrGc URUUUUUUUUU U S9 n U Sn URXS9n URU 5nU cSOURX5nU (dU4U SS-nUbU4U-$U$[ UUU R U R S9$)a^ labels (`tf.Tensor` 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). r,r$rNrrK)rrrjrJr rnr)r>rrrrrrrrrr.rrrr?rrs rIro*TFMobileBertForSequenceClassification.call$s:// ))%'/!5#"     ] F /~t4+?+?+OY,F)-)9TGf$ Ev E)!//))   rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=fNTrrj rsrtr8rurr\rvrjrfrrxs rIrv+TFMobileBertForSequenceClassification.build_ ::  4t , 8t334%%d+5 4t , 8t334%%tT4;;3J3J&KL54 95454C+.3C<+ C9< D rsrjrfrrrlrQ)rrrrrrrrrrrrrrrrrrr.rrrrOz"tuple | TFSequenceClassifierOutputrl)rPrQrRrSrR_keys_to_ignore_on_load_missingr^rr!rrr'_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATIONr r_SEQ_CLASS_EXPECTED_OUTPUT_SEQ_CLASS_EXPECTED_LOSSrorvrUr}r~s@rIrgrgs *& (2l# *+F+M+MNk+lm:.$2. .28<8<6:377;)-,0#'04 %0 *0 60 6 0 4 0 1 0 50 '0 *0 !0 .0 0  ,0 n0 d M MrLrgz MobileBert Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute `span start logits` and `span end logits`). c ^\rSrSr/SQrU4Sjr\\"\RS55\ "\ \ \ \\\\S9S S Sjj555rS SjrSrU=r$) TFMobileBertForQuestionAnsweringikrrrr\r>c>[TU]"U/UQ70UD6 URUl[USSS9Ul[ R RUR[UR5SS9Ul Xl g)NFrr@ qa_outputsr) r]r^rlrrrr_r`rrrrfrs rIr^)TFMobileBertForQuestionAnswering.__init__|ss 3&3F3 ++/%Vbc,,,,   /&BZBZ2[bn-  rLr)r r rqa_target_start_indexqa_target_end_indexrHrIc URUUUUUUUUU U S9 n U SnURU5n[R"USSS9unn[R"USS9n[R"USS9nSnU bU bXS.nUR UUU45nU (dUU4U SS-nUbU4U-$U$[ UUUU RU RS9$) a start_positions (`tf.Tensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. end_positions (`tf.Tensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. r,rrrrN)start_position end_position)r start_logits end_logitsrnr) rrr8splitsqueezerJr rnr)r>rrrrrrrrrstart_positions end_positionsrrrr?rrrr.rs rIro%TFMobileBertForQuestionAnswering.calls H// ))%'/!5#"  "!*1#%88FAB#? jzz,R8 ZZ 4   &=+D(7WF''z0JKD"J/'!"+=F)-)9TGf$ Ev E-%!!//))   rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=f)NTrr) rsrtr8rurr\rvrrfrrxs rIrv&TFMobileBertForQuestionAnswering.buildrvrw)rsrfrrlrr;)rrrrrrrrrrrrrrrrrrrrrrrrrOz&tuple | TFQuestionAnsweringModelOutputrl)rPrQrRrSrRr^rr!rrr_CHECKPOINT_FOR_QAr r_QA_TARGET_START_INDEX_QA_TARGET_END_INDEX_QA_EXPECTED_OUTPUT_QA_EXPECTED_LOSSrorvrUr}r~s@rIr~r~ks*&*+F+M+MNk+lm%2$40+'.28<8<6:377;)-,0#'9=7; %; *; 6; 6 ; 4 ; 1 ; 5; '; *; !; 7; 5; ;  0; n; z M MrLr~z MobileBert Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks. c^\rSrSr/SQrS/rU4Sjr\\"\ RS55\ "\ \ \S9S S Sjj555rS SjrS rU=r$) TFMobileBertForMultipleChoiceirhrc">[TU]"U/UQ70UD6 [USS9Ul[R R UR5Ul[R RS[UR5SS9Ul Xl g)Nrr[r$rjr)r]r^rrrr_rrrr`rrrjrfrs rIr^&TFMobileBertForMultipleChoice.__init__sy 3&3F3/\J||++F,F,FG ,,,, /&2J2J"KR^-  rLz(batch_size, num_choices, sequence_lengthrc Ub[U5Sn [U5Sn O[U5Sn [U5Sn Ub[R"USU 45OSnUb[R"USU 45OSnUb[R"USU 45OSnUb[R"USU 45OSnUb&[R"USU [U5S45OSnURUUUUUUUUU U S9 nUSnUR UU S9nUR U5n[R"USU 45nU cSOUR U U5nU (dU4USS-nUbU4U-$U$[UUURURS9$) a labels (`tf.Tensor` of shape `(batch_size,)`, *optional*): Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices]` where `num_choices` is the size of the second dimension of the input tensors. (See `input_ids` above) Nr$rrr)rrrrK) rr8r=rrrjrJr rnr)r>rrrrrrrrrr.r num_choices seq_lengthflat_input_idsflat_attention_maskflat_token_type_idsflat_position_idsflat_inputs_embedsrrr?reshaped_logitsrrs rIro"TFMobileBertForMultipleChoice.calls8  $Y/2K#I.q1J$]3A6K#M215JDMDYIJ/?@_cN\Nhbjj"j9IJnrN\Nhbjj"j9IJnrJVJbBJJ|b*5EFhl( JJ}r:z-7PQR7S&T U  //        #"     ]X F /**Vb+->?~t4+?+?+X%''!"+5F)-)9TGf$ Ev E*"!//))   rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=frsrtrxs rIrv#TFMobileBertForMultipleChoice.build?rvrw)rsrjrfrrrQ)rrrrrrrrrrrrrrrrrrr.rrrrOz#tuple | TFMultipleChoiceModelOutputrl)rPrQrRrSrRryr^rr!rrrrr rrorvrUr}r~s@rIrrs*& (2l#*#**+UV &/$.28<8<6:377;)-,0#'04 %? *? 6? 6 ? 4 ? 1 ? 5? '? *? !? .? ?  -? ? B M MrLrz MobileBert Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. c ^\rSrSr/SQrS/rU4Sjr\\"\ RS55\ "\ \ \\\S9S S Sjj555rS SjrS rU=r$) "TFMobileBertForTokenClassificationiKrrc>[TU]"U/UQ70UD6 URUl[USSS9UlUR b UR O UR n[RRU5Ul [RRUR[UR5SS9UlXlg)NFrr@rjrrkrns rIr^+TFMobileBertForTokenClassification.__init__]s 3&3F3 ++/%Vbc)/)B)B)NF % %TZTnTn ||++,>? ,,,,   /&BZBZ2[bn-  rLrrGc URUUUUUUUUU U S9 n U Sn URXS9n URU 5nU cSOURX5nU (dU4U SS-nUbU4U-$U$[ UUU R U R S9$)z labels (`tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`. r,rrNrrK)rrrjrJrrnr)r>rrrrrrrrrr.rrrr?rrs rIro'TFMobileBertForTokenClassification.callks6// ))%'/!5#"  "!*,,,J1~t4+?+?+OY,F)-)9TGf$ Ev E&!//))   rLcUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=frsrtrxs rIrv(TFMobileBertForTokenClassification.buildrvrwrxrQ)rrrrrrrrrrrrrrrrrrr.rrrrOztuple | TFTokenClassifierOutputrl)rPrQrRrSrRryr^rr!rrr$_CHECKPOINT_FOR_TOKEN_CLASSIFICATIONrr_TOKEN_CLASS_EXPECTED_OUTPUT_TOKEN_CLASS_EXPECTED_LOSSrorvrUr}r~s@rIrrKs *&(2l# *+F+M+MNk+lm7+$40 .28<8<6:377;)-,0#'04 %. *. 6. 6 . 4 . 1 . 5. '. *. !. .. .  ). n. ` M MrLr) r=rr[rr~rgrrrr)frT __future__rr% dataclassesrnumpynp tensorflowr8activations_tfrmodeling_tf_outputsrrr r r r r rmodeling_tf_utilsrrrrrrrrrrrrtf_utilsrrrutilsrrr r!r"r#configuration_mobilebertr% get_loggerrPloggerrrrrrrrrrrrzr{r|r)r_LayerrWLayerNormalizationrrrrrr rr,r6rArOr^rfrprrrrrrrrMOBILEBERT_START_DOCSTRINGrrrr=rr[rgr~rr__all__rMrLrIrs "!/       SR7   H %1$(L$l!='+E''!QQ8Mu||11M64%,,114 0u||!! 0% :e U\\//e Pb 2 2bJ+U\\//+B%3ELL..%3P++++8&,++&,R+ **+4,+5<<%%,+^+%,,$$+43##3.N& **N&b1&%,,,,1&hH++H@+%,,*<*<+>.5<<#5#5.b-%,,,,-$~(ELL..~(~(B%"3% /{/ /<(T5pj.,3., .,b d !<>Yd d NTVpq] 9;W] r] @Sell00S&ZV%,GIeV% V%r \M,GIe\M\M~ eM'BD[eMeMP gM$?AUgMgMT [M)DF_[M[M| rL