bCi/SrSSKJr SSKrSSKrSSKJr SSK J r J r J r J r JrJr SSKJrJrJrJrJrJrJrJrJrJrJrJr SSKJrJrJ r SS K!J"r"J#r#J$r$J%r% S S K&J'r' \%RP"\)5r*S r+S r,"SS\RZR\5r/"SS\RZR\5r0"SS\RZR\5r1"SS\RZR\5r2"SS\RZR\5r3"SS\RZR\5r4"SS\RZR\5r5"SS\RZR\5r6"SS\RZR\5r7"S S!\RZR\5r8\"S"S#\RZR\55r9"S$S%\5r:S&r;S'r<\#"S(\;5"S)S*\:55r="S+S,\RZR\5r>"S-S.\RZR\5r?\#"S/\;5"S0S1\:\55r@"S2S3\RZR\5rA\#"S4\;5"S5S6\:\55rB\#"S7\;5"S8S9\:\55rC\#"S:\;5"S;S<\:\55rD\#"S=\;5"S>S?\:\55rE/S@QrFg)AzTF 2.0 ConvBERT model.) annotationsN)get_tf_activation)TFBaseModelOutputTFMaskedLMOutputTFMultipleChoiceModelOutputTFQuestionAnsweringModelOutputTFSequenceClassifierOutputTFTokenClassifierOutput) TFMaskedLanguageModelingLossTFModelInputTypeTFMultipleChoiceLossTFPreTrainedModelTFQuestionAnsweringLossTFSequenceClassificationLossTFSequenceSummaryTFTokenClassificationLossget_initializerkeraskeras_serializable unpack_inputs)check_embeddings_within_bounds shape_liststable_softmax)add_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardlogging)ConvBertConfigzYituTech/conv-bert-baser cj^\rSrSrSrSU4SjjrSSjrS S SjjrSrU=r $) TFConvBertEmbeddings>zGConstruct the embeddings from word, position and token_type embeddings.c J>[TU]"S0UD6 XlURUlURUlUR Ul[ RRURSS9Ul [ RRURS9Ul g)N LayerNormepsilonname)rate)super__init__configembedding_sizemax_position_embeddingsinitializer_rangerlayersLayerNormalizationlayer_norm_epsr%Dropouthidden_dropout_probdropoutselfr-kwargs __class__s k/home/james-whalen/.local/lib/python3.13/site-packages/transformers/models/convbert/modeling_tf_convbert.pyr,TFConvBertEmbeddings.__init__As "6" $33'-'E'E$!'!9!988AVAV]h8i||++1K1K+L c[R"S5 URSURRUR /[ UR5S9UlSSS5 [R"S5 URSURRUR /[ UR5S9Ul SSS5 [R"S5 URSURUR /[ UR5S9Ul SSS5 UR(agSUl [USS5bf[R"URR 5 URR#SSURR /5 SSS5 gg!,(df  GNc=f!,(df  GN =f!,(df  N=f!,(df  g=f) Nword_embeddingsweight)r(shape initializertoken_type_embeddings embeddingsposition_embeddingsTr%)tf name_scope add_weightr- vocab_sizer.rr0r@type_vocab_sizerCr/rEbuiltgetattrr%r(buildr8 input_shapes r;rMTFConvBertEmbeddings.buildKs ]], -//{{--t/B/BC+D,B,BC*DK.]]2 3)-!{{22D4G4GH+D,B,BC*9*D &4]]0 1'+!33T5H5HI+D,B,BC(7(D $2 ::  4d + 7t~~223$$dD$++2L2L%MN43 81. -4 32 143s2A F=?A G'AG!3G2= G  G! G/2 HcUcUc [S5eUb>[XRR5 [R "UR US9n[U5SSnUc[R"USS9nUc-[R"[R"XWSU-S9SS 9n[R "URUS9n[R "URUS9n XH-U -n URU S 9n URXS 9n U $) zr Applies embedding based on inputs tensor. Returns: final_embeddings (`tf.Tensor`): output embedding tensor. Nz5Need to provide either `input_ids` or `input_embeds`.)paramsindicesr)dimsvaluer)startlimitaxis)inputs)r[training) ValueErrorrr-rIrFgatherr@rfill expand_dimsrangerErCr%r6) r8 input_ids position_idstoken_type_ids inputs_embedspast_key_values_lengthr\rOposition_embedstoken_type_embedsfinal_embeddingss r;callTFConvBertEmbeddings.callis  !6TU U  *9kk6L6L MIIT[[)LM /4  !WW+Q?N  >>5^Nd=delmL))4+C+C\ZIIT-G-GQ_`(:=NN>>1A>B<skQMO@'+)-+/*. & #& '& ) & ( & &  & & r=r"cB^\rSrSrU4SjrSrSSjrSSjrSrU=r $) TFConvBertSelfAttentionc >[TU]"S0UD6 URUR-S:wa&[ SURSURS35e[ URUR - 5nUS:aURUlSnOUnUR UlX@lURUlURUR-S:wa [ S5eURUR-UlURUR-Ul [RRUR[UR5SS9Ul[RRUR[UR5S S9Ul[RRUR[UR5S S9Ul[RR%URURS S[SUR- 5[UR5S S 9Ul[RRURUR-SS[UR5S9Ul[RRURSS[UR5S9Ul[RR-UR.5UlXlg)NrzThe hidden size (z6) is not a multiple of the number of attention heads ()rz6hidden_size should be divisible by num_attention_headsquerykernel_initializerr(keyrVsamekey_conv_attn_layer)padding activationdepthwise_initializerpointwise_initializerr(conv_kernel_layer)rr(rconv_out_layerr*)r+r, hidden_sizenum_attention_headsr]int head_ratioconv_kernel_sizeattention_head_size all_head_sizerr1Denserr0r}rrVSeparableConv1Drrrr4attention_probs_dropout_probr6r-)r8r-r9new_num_attention_headsrr:s r;r, TFConvBertSelfAttention.__init__s "6"    : : :a ?#F$6$6#78 445Q8  #&f&@&@6CTCT&T"U "Q &$88DO"# "9 $//DO#6 & 7 7    8 8 8A =UV V#)#5#59S9S#S !558P8PP\\''   ?6C[C[3\cj(  <<%%   ?6C[C[3\ch& \\''   ?6C[C[3\cj(  $)<<#?#?     ! !"1!d6K6K2K"L"1&2J2J"K&$@$  "'!3!3  $ $t'<'< <$.v/G/GH "4" $ll00   !.v/G/GH 1 ||++F,O,OP  r=c[R"XSURUR45n[R"U/SQS9$)NrTrrrperm)rFreshaperr transpose)r8x batch_sizes r;transpose_for_scores,TFConvBertSelfAttention.transpose_for_scoress6 JJqr4+C+CTE]E]^ _||AL11r=c [U5SnURU5nURU5nURU5n UR U5n UR Xv5n UR X5n [ R"X5n URU 5n[ R"USURS/5n[USS9n[ R"SS/[URS- S- 5[URS- S- 5/SS//5nURU5n[ R"UUSUR/5n[ R "UUS5n[ R""[%UR5Vs/sH7n[ R&"USUS/U[U5SUR/5PM9 snSS9n[ R"USUR(UR/5n[ R*"UU5n[ R"USUR/5n[ R*"XSS9n[ R,"[U 5SUR.5nU[ R0R3U5- nUbUU-n[USS9nUR5UUS 9nUbUU-n[ R"XSUR6UR(/5n[ R8"U/S Q5n[ R*"UU5n[ R8"U/S QS 9n[ R"UUSUR6UR(/5n[ R:"UU/S5n[ R"UUSUR<UR-45nU(aUU4nU$U4nU$s snf) NrrTrrYrCONSTANTT) transpose_br\rr)rr}rrVrrrFmultiplyrrrrconstantrrrpadstackraslicermatmulcastdtypemathsqrtr6rrconcatr)r8 hidden_statesattention_mask head_maskoutput_attentionsr\rmixed_query_layermixed_key_layermixed_value_layermixed_key_conv_attn_layer query_layer key_layerconv_attn_layerrpaddingsriunfold_conv_out_layerattention_scoresdkattention_probs value_layer context_layerconv_outoutputss r;rjTFConvBertSelfAttention.calls .q1  JJ}5((=1 JJ}5$($<$<]$K!//0AN --oJ ++&?S 22?CJJ'82t?T?TVW:XY*+<1E;;d++a/145sD3DENR9K9K4LM99  WWZ *2.0@0F0F G+bggll2.>>  %/.@ ))9C,,,J  - 9Ojj B0H0H$JbJbc ll; =  /;?  ]F ::nz2t?W?WY]YqYq.rs =(";Q?  JDOOd>P>P,PQ 7H=/2O\M]c s>O c\UR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5b[[R"URR 5 URR SSUR/5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  GNF=f!,(df  GN=f!,(df  GN=f!,(df  GN#=f!,(df  N=f!,(df  g=f)NTr}rrVrrr)rKrLrFrGr}r(rMr-rrrVrrrrrNs r;rMTFConvBertSelfAttention.build)s" ::  4$ ' 3tzz/   $dkk.E.E!FG0 4 % 1txx}}-dDKK,C,CDE. 4$ ' 3tzz/   $dkk.E.E!FG0 4. 5 At77<<=((..dDKK 4,d 3 ?t55::;&&,,dD$:L:L-MN< 4)4 0 <t22778##))4t{{7N7N*OP98 =0/.-0/>=<;98sH3K3K83K(+3K:)L 3L K K%( K7: L  L L+)rrrKr-rrrr6rrrrr}rVFrl) rprqrrrsr,rrjrMrurvrws@r;ryrys!<|2 QfQQr=ryc<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFConvBertSelfOutputiAc j>[TU]"S0UD6 [RR UR [ UR5SS9Ul[RRURSS9Ul [RRUR5UlXlgNdenser~r%r&r*)r+r,rr1rrrr0rr2r3r%r4r5r6r-r7s r;r,TFConvBertSelfOutput.__init__Bs "6"\\''   ?6C[C[3\cj(  88AVAV]h8i||++F,F,FG  r=clURU5nURXS9nURX-5nU$Nrrr6r%r8r input_tensorr\s r;rjTFConvBertSelfOutput.callL7 =1  ] F }'CD 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% rKrLrFrGrr(rMr-rr%rNs r;rMTFConvBertSelfOutput.buildS ::  4$ ' 3tzz/   $dkk.E.E!FG0 4d + 7t~~223$$dD$++2I2I%JK43 80/433D3D D D!r%rKr-rr6rrl rprqrrrsr,rjrMrurvrws@r;rrAs L Lr=rcB^\rSrSrU4SjrSrSSjrSSjrSrU=r $) TFConvBertAttentioni_c b>[TU]"S0UD6 [USS9Ul[ USS9Ulg)Nr8r(outputr*)r+r,ryself_attentionr dense_outputr7s r;r,TFConvBertAttention.__init__`s1 "6"5f6J0hGr=c[erlNotImplementedError)r8headss r; prune_headsTFConvBertAttention.prune_headsfs!!r=c`URXX4US9nURUSXS9nU4USS-nU$Nrrr)rr) r8rrrrr\ self_outputsattention_outputrs r;rjTFConvBertAttention.callisT** )QY+  ,,\!_l,^#% 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)rKrLrFrGrr(rMrrNs r;rMTFConvBertAttention.buildrs ::  4)4 0 <t22778##))$/9 4 . :t00556!!''-76 ;9876C.C% C"% C3)rKrrrrl) rprqrrrsr,rrjrMrurvrws@r;rr_sH " . .r=rc>^\rSrSrU4SjrSU4SjjrSrSrU=r$)GroupedLinearLayeri~c >[TU]"S0UD6 XlX lX0lX@lURUR-UlURUR-UlgNr*)r+r, input_size output_size num_groupsr group_in_dim group_out_dim)r8rrrrr9r:s r;r,GroupedLinearLayer.__init__sU "6"$&$"4 OOt>!--@r=c>URSURURUR/URSS9UlURSUR /URURSS9Ul[TU])U5 g)NkernelT)rArB trainablebias)rArBrr) rHrrrrrrrrr+rMr8rOr:s r;rMGroupedLinearLayer.buildsoo %%t'8'8$//J// &  OO 4++,$:Q:QY]YcYcos$    k"r=c[U5Sn[R"[R"USURUR /5/SQ5n[R "U[R"UR/SQ55n[R"U/SQ5n[R"X2SUR/5n[RRX0RS9nU$)NrrT)rrr)rrrrVr) rrFrrrrrrrnnbias_addr)r8rrrs r;rjGroupedLinearLayer.calls .q1 LLMBIZIZ3[\^g h IIadkk9= > LLI & JJqr4+;+;< = EENNN 3r=)rrrrrrrrrl) rprqrrrsr,rMrjrurvrws@r;rr~sA #r=rc8^\rSrSrU4SjrSrSSjrSrU=r$)TFConvBertIntermediateic >[TU]"S0UD6 URS:XaB[RR UR [UR5SS9Ul OC[URUR UR[UR5SS9Ul [UR[5(a[UR5UlOURUlXlg)Nrrr~rrr(r*)r+r,rrr1rintermediate_sizerr0rrr isinstance hidden_actstrrintermediate_act_fnr-r7s r;r,TFConvBertIntermediate.__init__s "6"    !++((_VMeMe=fmt,DJ,""((!,,#263K3K#L DJ f'' - -'89J9J'KD $'-'8'8D $ r=cJURU5nURU5nU$rl)rrr8rs r;rjTFConvBertIntermediate.calls& =1 00? r=c@UR(agSUl[USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  g=f)NTr) rKrLrFrGrr(rMr-rrNs r;rMTFConvBertIntermediate.buildsm ::  4$ ' 3tzz/   $dkk.E.E!FG0/ 4//s 3B B)rKr-rrrlrrws@r;r r s* HHr=r c<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFConvBertOutputic >[TU]"S0UD6 URS:XaB[RR UR [UR5SS9Ul OC[URUR UR[UR5SS9Ul [RRURSS9Ul[RRUR 5UlXlg)Nrrr~r r%r&r*)r+r,rrr1rrrr0rrr r2r3r%r4r5r6r-r7s r;r,TFConvBertOutput.__init__s "6"    !++""vG_G_7`gn,DJ,((""!,,#263K3K#L DJ88AVAV]h8i||++F,F,FG  r=clURU5nURXS9nURX-5nU$rrrs r;rjTFConvBertOutput.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=fNTr%r) rKrLrFrGr%r(rMr-rrr rNs r;rMTFConvBertOutput.builds ::  4d + 7t~~223$$dD$++2I2I%JK4 4$ ' 3tzz/   $dkk.K.K!LM0/ 4430/rrrrlrrws@r;rrs& N Nr=rc<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFConvBertLayeric >[TU]"S0UD6 [USS9Ul[ USS9Ul[ USS9Ulg)N attentionr intermediaterr*)r+r,rr#r r$r bert_outputr7s r;r,TFConvBertLayer.__init__s? "6",V+F26O+FBr=cURXX4US9nUSnURU5nURXUS9n U 4USS-n U $r)r#r$r%) r8rrrrr\attention_outputsrintermediate_output layer_outputrs r;rjTFConvBertLayer.callsm NN 9RZ+ -Q/"//0@A''(;X`'a /$5ab$99r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  N=f!,(df  N}=f!,(df  g=f)NTr#r$r%) rKrLrFrGr#r(rMr$r%rNs r;rMTFConvBertLayer.builds ::  4d + 7t~~223$$T*4 4 . :t00556!!''-7 4 - 9t//445  &&t,65 : 437665$D0.E E0 D> E E )r#r%rKr$rrlrrws@r;r!r!sC  - -r=r!c>^\rSrSrU4SjrSSjrSSjrSrU=r$)TFConvBertEncoderi c >[TU]"S0UD6 [UR5Vs/sHn[ USU3S9PM snUlgs snf)Nzlayer_._rr*)r+r,ranum_hidden_layersr!layer)r8r-r9rr:s r;r,TFConvBertEncoder.__init__sG "6"LQRXRjRjLklLkqofXaS>BLkl lsA c $U(aSOSnU(aSOSn [UR5H1upU(aX4-nU "XX:XGS9n U SnU(dM)XS4-n M3 U(aX4-nU(d[SXU 455$[XU S9$)Nr*rrrc3.# UH ocMUv M g7frlr*).0vs r; )TFConvBertEncoder.call..1sh$Vq$Vs )last_hidden_stater attentions) enumerater3tupler) r8rrrroutput_hidden_states return_dictr\all_hidden_statesall_attentionsr layer_module layer_outputss r;rjTFConvBertEncoder.calls#7BD0d(4OA#$58H$H!(y|=NM*!,M  !/3C2E!E 5 14D D h]~$Vhh h +Yg  r=cUR(agSUl[USS5bNURH=n[R"UR 5 UR S5 SSS5 M? gg!,(df  MR=f)NTr3)rKrLr3rFrGr(rM)r8rOr3s r;rMTFConvBertEncoder.build7s` ::  4$ ' 3]]5::.KK%/.$ 4..s A77 B )rKr3rrlrrws@r;r0r0 sm" H&&r=r0c8^\rSrSrU4SjrSrSSjrSrU=r$)!TFConvBertPredictionHeadTransformiAc >[TU]"S0UD6 [RR UR [ UR5SS9Ul[UR[5(a[UR5Ul OURUl [RRURSS9UlXlgr)r+r,rr1rr.rr0rr rrrtransform_act_fnr2r3r%r-r7s r;r,*TFConvBertPredictionHeadTransform.__init__Bs "6"\\''  ! !ofF^F^6_fm(   f'' - -$5f6G6G$HD !$*$5$5D !88AVAV]h8i r=clURU5nURU5nURU5nU$rl)rrKr%rs r;rj&TFConvBertPredictionHeadTransform.callQs4 =1 --m< }5 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=frrrNs r;rM'TFConvBertPredictionHeadTransform.buildXrr)r%rKr-rrKrlrrws@r;rIrIAs  L Lr=rIc|^\rSrSr\rU4SjrSrSrSr Sr Sr \ S Sj5r S S jrS rU=r$) TFConvBertMainLayeridc >[TU]"S0UD6 [USS9UlURUR :wa-[ RRUR SS9Ul [USS9Ul Xl g)NrDrembeddings_projectencoderr*) r+r,r"rDr.rrr1rrTr0rUr-r7s r;r,TFConvBertMainLayer.__init__hsk "6".vLI  F$6$6 6&+ll&8&89K9KRf&8&gD #(i@  r=cUR$rl)rDr8s r;get_input_embeddings(TFConvBertMainLayer.get_input_embeddingsss r=c`XRlURSURlgNr)rDr@rArIr8rVs r;set_input_embeddings(TFConvBertMainLayer.set_input_embeddingsvs !&%*[[^"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 TFConvBertMainLayer._prune_headszs "!r=cUc[R"US5n[R"XSSSUS45n[R"XC5nSU- S-nU$)Nrrg?g)rFr_rr)r8rrOrextended_attention_masks r;get_extended_attention_mask/TFConvBertMainLayer.get_extended_attention_masksg  !WW[!4N#%**^!naQRT_`aTb=c"d#%''*A"I#&)@#@H"L&&r=cLUb[eS/URR-nU$rl)rr-r2)r8rs r; get_head_mask!TFConvBertMainLayer.get_head_masks*  % %!>!>>Ir=c UbUb [S5eUb [U5n OUb[U5SSn O [S5eUc[R"U S5nUc[R"U S5nUR XX6U S9n UR X+U R 5n URU5n[US5(aURXS9n URU U UUUU U S9n U $)NzDYou cannot specify both input_ids and inputs_embeds at the same timerTz5You have to specify either input_ids or inputs_embedsrrrrT) r]rrFr_rDrfrrihasattrrTrU)r8rbrrdrcrrerr?r@r\rOrres r;rjTFConvBertMainLayer.calls   ]%>cd d  "$Y/K  &$]3CR8KTU U  !WW[!4N  !WW[!4N iqr "&"B"B>`m`s`s"t&&y1 4- . . 33M3UM  #   % r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  N=f!,(df  g=f)NTrDrUrT) rKrLrFrGrDr(rMrUrTr-r.rNs r;rMTFConvBertMainLayer.builds ::  4t , 8t334%%d+5 4D ) 5t||001 ""4(2 4-t 4 @t66;;<''--tT4;;;U;U.VW=< A 5421=<$E.E 3E) E E&) E7)rKr-rDrTrU NNNNNNNNNFrl)rprqrrrsr config_classr,rYr^rbrfrirrjrMrurvrws@r;rRrRdsg!L 4"'*!--^ X Xr=rRc \rSrSrSr\rSrSrg)TFConvBertPreTrainedModelizz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. convbertr*N) rprqrrrsrtr rrbase_model_prefixrur*r=r;rtrts "L"r=rtax 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 ([`ConvBertConfig`]): 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). zbThe bare ConvBERT Model transformer outputting raw hidden-states without any specific head on top.c^\rSrSrU4Sjr\\"\RS55\ "\ \ \ S9SS Sjj555r S SjrSrU=r$) TFConvBertModeliIcL>[TU]"U/UQ70UD6 [USS9Ulg)Nrur)r+r,rRrur8r-r[r9r:s r;r,TFConvBertModel.__init__Ns( 3&3F3+FD r=batch_size, sequence_length checkpoint output_typerrc 6URUUUUUUUUU U S9 n U $)N rbrrdrcrrerr?r@r\)ru) r8rbrrdrcrrerr?r@r\rs r;rjTFConvBertModel.callSs<(--))%'/!5#  r=cUR(agSUl[USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  g=f)NTru)rKrLrFrGrur(rMrNs r;rMTFConvBertModel.buildvs^ ::  4T * 6t}}112 ##D)32 722s A88 B)rKrurq)rbTFModelInputType | Nonernp.array | tf.Tensor | Nonerdrrcrrrrermr bool | Noner?rr@rr\rnroz$TFBaseModelOutput | tuple[tf.Tensor]rl)rprqrrrsr,rrCONVBERT_INPUTS_DOCSTRINGformatr_CHECKPOINT_FOR_DOCr_CONFIG_FOR_DOCrjrMrurvrws@r;rxrxIs E *+D+K+KLi+jk&%$.26:6:4815*.)-,0#'*44  2  / ('*! . l8**r=rxcR^\rSrSrU4SjrU4SjrSrSrSrSr Sr S r U=r $) TFConvBertMaskedLMHeadic `>[TU]"S0UD6 XlURUlX lgr)r+r,r-r.input_embeddings)r8r-rr9r:s r;r,TFConvBertMaskedLMHead.__init__s, "6" $33 0r=c|>URURR4SSSS9Ul[TU]U5 g)NzerosTr)rArBrr()rHr-rIrr+rMrs r;rMTFConvBertMaskedLMHead.builds7OO4;;+A+A*CQXdhouOv   k"r=cUR$rl)rrXs r;get_output_embeddings,TFConvBertMaskedLMHead.get_output_embeddingss$$$r=c^XRl[U5SURlgr\)rr@rrIr]s r;set_output_embeddings,TFConvBertMaskedLMHead.set_output_embeddingss&',$+5e+r=rc<^\rSrSrU4SjrSSjrSSjrSrU=r$)TFConvBertGeneratorPredictionsic >[TU]"S0UD6 [RR UR SS9Ul[RRURSS9Ul Xl g)Nr%r&rrr*) r+r,rr1r2r3r%rr.rr-r7s r;r,'TFConvBertGeneratorPredictions.__init__s[ "6"88AVAV]h8i\\''(=(=G'L  r=clURU5n[S5"U5nURU5nU$)Ngelu)rrr%)r8generator_hidden_statesr\rs r;rj#TFConvBertGeneratorPredictions.calls4 #:; )&1-@ }5 r=cHUR(agSUl[USS5be[R"URR 5 URR SSURR/5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  g=fr) rKrLrFrGr%r(rMr-r.rrrNs r;rM$TFConvBertGeneratorPredictions.builds ::  4d + 7t~~223$$dD$++2L2L%MN4 4$ ' 3tzz/   $dkk.E.E!FG0/ 4430/r)r%rKr-rrrlrrws@r;rrs H Hr=rz6ConvBERT Model with a `language modeling` head on top.c^\rSrSrU4SjrSrSr\\"\ RS55\ "\ \ \S9S S Sjj555rS SjrS rU=r$) TFConvBertForMaskedLMicL>[TU]"U40UD6 Xl[USS9Ul[ USS9Ul[UR[5(a[UR5Ul OURUl [XRRSS9Ulg)Nrurgenerator_predictionsgenerator_lm_head)r+r,r-rRrurrr rrrrrrDrrzs r;r,TFConvBertForMaskedLM.__init__s *6* +FD %CFQh%i" f'' - -/0A0ABDO$//DO!7 @X@X_r!sr=cUR$rl)rrXs r; get_lm_head!TFConvBertForMaskedLM.get_lm_heads%%%r=cNURS-URR-$)N/)r(rrXs r;get_prefix_bias_name*TFConvBertForMaskedLM.get_prefix_bias_names!yy3!7!7!! 667P6d 223D2X~t4+?+?+Z'),CAB,GGF)-)9TGf$ Ev E$1??.99   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bO[R"URR 5 URR S5 SSS5 gg!,(df  N=f!,(df  N}=f!,(df  g=f)NTrurr) rKrLrFrGrur(rMrrrNs r;rMTFConvBertForMaskedLM.build s ::  4T * 6t}}112 ##D)3 40$ 7 Ct99>>?**006@ 4,d 3 ?t55::;&&,,T2<; @ 32@?<;r.)rrKr-rurr NNNNNNNNNNF)rbrrnp.ndarray | tf.Tensor | Nonerdrrcrrrrermrrr?rr@rrrmr\rroztuple | TFMaskedLMOutputrl)rprqrrrsr,rrrrrrrrrrrjrMrurvrws@r;rrs t&=*+D+K+KLi+jk&$$.28<8<6:37*.)-,0#'#' %/ */ 6/ 6 / 4 / 1 / (/ '/ */ !/ !/ /  "/ l/ b 3 3r=rc<^\rSrSrSrU4SjrSrSSjrSrU=r $)TFConvBertClassificationHeadiz-Head for sentence-level classification tasks.c >[TU]"S0UD6 [RR UR [ UR5SS9UlURb URO URn[RRU5Ul [RR UR[ UR5SS9UlXlg)Nrr~out_projr*)r+r,rr1rrrr0rclassifier_dropoutr5r4r6 num_labelsrr-)r8r-r9rr:s r;r,%TFConvBertClassificationHead.__init__s "6"\\''   ?6C[C[3\cj(  *0)B)B)NF % %TZTnTn ||++,>?  **   /&BZBZ2[bl+   r=c USS2SSS24nURU5nURU5n[URR5"U5nURU5nUR U5nU$r\)r6rrr-rr)r8rr9rs r;rj!TFConvBertClassificationHead.call/se !Q' " LLO JJqM dkk44 5a 8 LLO MM! 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=f)NTrr) rKrLrFrGrr(rMr-rrrNs r;rM"TFConvBertClassificationHead.build9s ::  4$ ' 3tzz/   $dkk.E.E!FG0 4T * 6t}}112 ##T41H1H$IJ32 70/32r)rKr-rr6rrl) rprqrrrsrtr,rjrMrurvrws@r;rrs7  K Kr=rzp ConvBERT Model transformer with a sequence classification/regression head on top e.g., for GLUE tasks. c^\rSrSrU4Sjr\\"\RS55\ "\ \ \ S9SS Sjj555r S SjrSrU=r$) #TFConvBertForSequenceClassificationiEc>[TU]"U/UQ70UD6 URUl[USS9Ul[ USS9Ulg)Nrur classifier)r+r,rrRrurrrzs r;r,,TFConvBertForSequenceClassification.__init__LsC 3&3F3 +++FD 6vLQr=r|r}c URUUUUUUUUU U S9 n URU SU S9n U cSOURX5nU (dU 4U SS-nUbU4U-$U$[UU U RU R S9$)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). rrdrcrrerr?r@r\rrNrr)rurrr rr<)r8rbrrdrcrrerr?r@rr\rrrrs r;rj(TFConvBertForSequenceClassification.callRs6-- ))%'/!5#  h?~t4+?+?+OY,F)-)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=fNTrur)rKrLrFrGrur(rMrrNs r;rM)TFConvBertForSequenceClassification.builds ::  4T * 6t}}112 ##D)3 4t , 8t334%%d+54 93254r)rKrrurr)rbrrrrdrrcrrrrermrrr?rr@rrrmr\rroz"tuple | TFSequenceClassifierOutputrl)rprqrrrsr,rrrrrrr rrjrMrurvrws@r;rrEsR *+D+K+KLi+jk&.$.28<8<6:37*.)-,0#'#' %- *- 6- 6 - 4 - 1 - (- '- *- !- !- -  ,- l- ^ , ,r=rz ConvBERT Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks. c^\rSrSrU4Sjr\\"\RS55\ "\ \ \ S9SS Sjj555r S SjrSrU=r$) TFConvBertForMultipleChoiceic>[TU]"U/UQ70UD6 [USS9Ul[ XR SS9Ul[RRS[UR 5SS9Ul Xl g)Nrursequence_summary)r0r(rrr~) r+r,rRrurr0rrr1rrrr-rzs r;r,$TFConvBertForMultipleChoice.__init__sx 3&3F3+FD 1 &>&>EW!  ,,,, /&2J2J"KR^-  r=z(batch_size, num_choices, sequence_lengthr}c Ub[U5Sn [U5Sn O[U5Sn [U5Sn Ub[R"USU 45OSnUb[R"USU 45OSnUb[R"USU 45OSnUb[R"USU 45OSnUb&[R"USU [U5S45OSnURUUUUUUUUU U S9 nUR USU S9nUR U5n[R"USU 45nU cSOUR U U5nU (dU4USS-nUbU4U-$U$[UUURURS 9$) 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) NrrrTr)r@r\rrr) rrFrrurrrrrr<)r8rbrrdrcrrerr?r@rr\ num_choicesrflat_input_idsflat_attention_maskflat_token_type_idsflat_position_idsflat_inputs_embedsrrreshaped_logitsrrs r;rj TFConvBertForMultipleChoice.calls8  $Y/2K#I.q1J$]3A6K#M215JDMDYIJ/?@_cN\Nhbjj"j9IJnrN\Nhbjj"j9IJnrJVJbBJJ|b*5EFhl( JJ}r:z-7PQR7S&T U  --        #  &&wqzH&E(**Vb+->?~t4+?+?+X%''!"+5F)-)9TGf$ Ev E*"!//))   r=cUR(agSUl[USS5bN[R"URR 5 URR S5 SSS5 [USS5bN[R"URR 5 URR S5 SSS5 [USS5bf[R"URR 5 URR SSURR/5 SSS5 gg!,(df  N=f!,(df  N=f!,(df  g=f)NTrurr) rKrLrFrGrur(rMrrr-rrNs r;rM!TFConvBertForMultipleChoice.builds ::  4T * 6t}}112 ##D)3 4+T 2 >t4499:%%++D1; 4t , 8t334%%tT4;;3J3J&KL54 9 32;:54rp)rKrr-rurr)rbrrrrdrrcrrrrermrrr?rr@rrrmr\rroz#tuple | TFMultipleChoiceModelOutputrl)rprqrrrsr,rrrrrrrrrjrMrurvrws@r;rrs *!(()ST &/$.28<8<6:37*.)-,0#'#' %> *> 6> 6 > 4 > 1 > (> '> *> !> !> >  -> > @ M Mr=rz ConvBERT Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. c^\rSrSrU4Sjr\\"\RS55\ "\ \ \ S9SS Sjj555r S SjrSrU=r$) TFConvBertForTokenClassificationic>[TU]"U/UQ70UD6 URUl[USS9UlUR b UR O UR n[RRU5Ul [RRUR[UR5SS9UlXlg)Nrurrr~)r+r,rrRrurr5rr1r4r6rrr0rr-)r8r-r[r9rr:s r;r,)TFConvBertForTokenClassification.__init__s 3&3F3 +++FD )/)B)B)NF % %TZTnTn ||++,>? ,,,,   /&BZBZ2[bn-  r=r|r}c URUUUUUUUUU U S9 n U Sn URXS9n URU 5nU cSOURX5nU (dU4U SS-nUbU4U-$U$[ UUU R U R S9$)z labels (`tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`. rrrNrr)rur6rrr rr<)r8rbrrdrcrrerr?r@rr\rsequence_outputrrrs r;rj%TFConvBertForTokenClassification.calls2-- ))%'/!5#  "!*,,,J1~t4+?+?+OY,F)-)9TGf$ 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=fr) rKrLrFrGrur(rMrr-rrNs r;rM&TFConvBertForTokenClassification.buildK ::  4T * 6t}}112 ##D)3 4t , 8t334%%tT4;;3J3J&KL54 93254C+.3C<+ C9< D )rKrr-rur6rr)rbrrrrdrrcrrrrermrrr?rr@rrrmr\rroztuple | TFTokenClassifierOutputrl)rprqrrrsr,rrrrrrr rrjrMrurvrws@r;rrs *+D+K+KLi+jk&+$.28<8<6:37*.)-,0#'#' %, *, 6, 6 , 4 , 1 , (, ', *, !, !, ,  ), l, \ M Mr=rz ConvBERT Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layer on top of the hidden-states output to compute `span start logits` and `span end logits`). c^\rSrSrU4Sjr\\"\RS55\ "\ \ \ S9SS Sjj555r S SjrSrU=r$) TFConvBertForQuestionAnsweringiWc>[TU]"U/UQ70UD6 URUl[USS9Ul[ R RUR[UR5SS9Ul Xl g)Nrur qa_outputsr~) r+r,rrRrurr1rrr0rr-rzs r;r,'TFConvBertForQuestionAnswering.__init___sp 3&3F3 +++FD ,,,,   /&BZBZ2[bn-  r=r|r}c URUUUUUUUUU U S9 n U SnURU5n[R"USSS9unn[R"USS9n[R"USS9nSnU b U bSU 0nU US'UR UUU45nU (dUU4U S S-nUbU4U-$U$[ UUUU RU RS 9$) a start_positions (`tf.Tensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. end_positions (`tf.Tensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. rrrrTrYNstart_position end_positionr)r start_logits end_logitsrr<) rurrFsplitsqueezerr rr<)r8rbrrdrcrrerr?r@start_positions end_positionsr\rrrrrrrrs r;rj#TFConvBertForQuestionAnswering.callis@-- ))%'/!5#  "!*1#%88FAB#? jzz,R8 ZZ 4   &=+D&8F%2F> "''z0JKD"J/'!"+=F)-)9TGf$ 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)NTrur) rKrLrFrGrur(rMrr-rrNs r;rM$TFConvBertForQuestionAnswering.buildrr)rKr-rurr) NNNNNNNNNNNF)rbrrrrdrrcrrrrermrrr?rr@rrrmrrmr\rroz&tuple | TFQuestionAnsweringModelOutputrl)rprqrrrsr,rrrrrrr rrjrMrurvrws@r;rrWs*+D+K+KLi+jk&2$.28<8<6:37*.)-,0#',0*. %; *; 6; 6 ; 4 ; 1 ; (; '; *; !; *; (; ;  0; l; z M Mr=r)rrrrrr!rxrt)Grt __future__rnumpynp tensorflowrFactivations_tfrmodeling_tf_outputsrrrr r r modeling_tf_utilsr r rrrrrrrrrrtf_utilsrrrutilsrrrrconfiguration_convbertr get_loggerrploggerrrr1Layerr"ryrrrr rr!r0rIrRrtCONVBERT_START_DOCSTRINGrrxrrrrrrrr__all__r*r=r;rs"/    SR 3   H %/"Q 5<<--Q hlQell00lQ^L5<<--L<.%,,,,.>++B"HU\\//"HJ$Nu||))$NN-ell((-D1& **1&h L (:(: LFuX%,,,,uXuXp# 1#(T5ph/*//* /*d"U\\//"JHU\\%7%7H6RTlmY357SY3nY3x&K5<<#5#5&KR  F,*CEaF,  F,R bM";=QbMbMJ MM'@B[MMMM` XM%>@WXMXMv r=