+ j9HRt^RIt^RIt^RIHt^RIt^RIHt^RIH t ^RI H t ^RI Ht^RIHt^RIHt^R IHtHtHtHtHtHt^R IHt^R IHtHt^R IH t ]PB!]"4t#]!R R7]!RR]444t$!RR]PJ4t&!RR]PJ4t'!RR]PJ4t(!RR]PJ4t)!RR]PJ4t*!RR]PJ4t+!RR]PJ4t,!RR ]PJ4t-!R!R"]4t.!R#R$]PJ4t/]!R%R&]44t0]!R'R(]044t1!R)R*]PJ4t2]!R+R7!R,R-]044t3!R.R/]PJ4t4!R0R1]PJ4t5]!R2R7!R3R4]044t6]!R5R7!R6R7]044t7]!R8R7!R9R:]044t8]!R;R<]044t9.R=Ot:R#)>zPyTorch ViLT model.N) dataclass)nn)CrossEntropyLoss)initialization)ACT2FN)create_bidirectional_mask)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPoolingMaskedLMOutput ModelOutputSequenceClassifierOutputTokenClassifierOutput)PreTrainedModel)auto_docstringlogging ViltConfigzF Class for outputs of [`ViltForImagesAndTextClassification`]. ) custom_introc@a]tRt^,toRtRtRtRtRtV3Rlt Rt Vt R#)(ViltForImagesAndTextClassificationOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Classification (or regression if config.num_labels==1) loss. logits (`torch.FloatTensor` of shape `(batch_size, config.num_labels)`): Classification (or regression if config.num_labels==1) scores (before SoftMax). hidden_states (`list[tuple(torch.FloatTensor)]`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): List of tuples of `torch.FloatTensor` (one for each image-text pair, each tuple containing 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. Nc<V^8dQh/S[PR,;R&S[PR,;R&S[S[S[P,,R,;R&S[S[S[P,,R,;R&#)Nlosslogits hidden_states attentions)torch FloatTensorlisttuple)format __classdict__s"g/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/vilt/modeling_vilt.py __annotate__5ViltForImagesAndTextClassificationOutput.__annotate__,sy$   d ")%&    $+'(e//01D8?)*U5,,-.5<+) __name__ __module__ __qualname____firstlineno____doc__rrrr__annotate_func____static_attributes____classdictcell__)r"s@r#rr,s' &*D'+F;?M8 projectionweightshapefloatr functional interpolatelongsumrG image_size patch_sizer@ transposeviewrcatzippadflattenstackmeshgridarangetorVexpand isinstanceintmaxminnonzerouniquerM enumerate multinomialonesappendr<rF)'rH pixel_values pixel_maskmax_image_length_phpwxx_maskx_hx_w batch_size num_channelsheightwidth patch_dim spatial_poshw pos_embed patch_indexeffective_resolution valid_idx non_valid_idx unique_rowsu valid_row_idxnon_valid_row_idxv valid_numsnon_valid_numspad_numsselectinvp valid_choice pad_choice cls_tokenss'&&&& r# visual_embedViltEmbeddings.visual_embed\s,,77>>DD 1"  ! !, /M*002**6QWWQZ8P*QVVXTl1%d+Tl1%d+23''/ &KK**dkk.D.DD ..x8BB1aHMMaQ]jst II M *DAq !!MM--#V'&* .  1fqj1*  %%a(221a8 IIaL " "1a (kk NN5<< R(895<< UWHX;Ydh ioq "FMM" " ""56 !((a&,,q/2rSUV !))!Q/ " a #3#;:N^`cCdCd $'9 3779 #&9 "#7#;#;#=?OP NNEN2 V,,e,< o,,. BMN+Q9T?a#788+ NNYZk]=+>!+CDDkZ)67AaffQi 7->?->!&&)->?2<=*Q$q((*=&s:~x'PQMAz2qAv$00A1D1D1FHXY  mA.|<="..uzz"~/C/C/EqVZ[  eiiq)9;LQ;OPZ;[(\bcde R6q) fTlF4L( ) . .z2| Lt fTl2388RH!&,t "<=BB:rSTU fTlF4L89>>z2|\ ^^**:r2> IIz1o1 -II  % %g .z : A A*bRT UW` agh   M LLOEJJv||A:==fEvNTUV;888S POZ7?=s%;A!\ \"\ \$\\c 6VPWVR7p Vf,VPWEVPPR7wrzp MVP ^4p Vf^pWP \ P!V\ PV PR74,p WpP \ P!W\ PV PR74,p\ P!W.^R7p \ P!W*.^R7p W3#)) input_idstoken_type_ids inputs_embeds)rdtyperVrN) r8rrGrrmrCr zeros_likerdrV full_likerj)rHrattention_maskrr}r~r image_embedsimage_token_type_idx text_embeds image_masksr embeddingsmaskss&&&&&&&&& r#forwardViltEmbeddings.forwards**m+  595F5F4;;;W;W6G6 2L{%,,Q/K '#$ !$>$>   ^5::kFXFX Y%  $&@&@ OOKUZZXcXjXj k'  YY :B  >7Q?  r&)r<rGrFr>r@r8rC))r4) r(r)r*r+r,r6rrr.r/ __classcell__rIr"s@@r#r1r1Ds  V9p'!'!r&r1c@aa]tRt^toRtV3RltRRltRtVtV;t #)r7zGConstruct the embeddings from word, position and token_type embeddings.cN<\SV`4\P!VPVP VP R7Vn\P!VPVP 4Vn \P!VPVP 4Vn \P!VP VPR7Vn \P!VP4Vn\P"!\$P&!VP4P)R4RR7Vn\P"!\$P,!VP*P/4\$P0R7RR7VnR#)) padding_idxepsF) persistentrNr4r])r5r6rrA vocab_sizer; pad_token_idword_embeddingsmax_position_embeddingsr@type_vocab_sizerC LayerNormlayer_norm_epsrDrErFBufferrrprr position_idsr:rMrdrrHrGrIs&&r#r6TextEmbeddings.__init__s  !||F,=,=v?Q?Q_e_r_rs#%<<0N0NPVPbPb#c %'\\&2H2H&J\J\%]"f&8&8f>S>STzz&"<"<= IIell63Q3Q&R&Y&YZa&botu ii D4E4E4J4J4LTYT^T^(_lqrr&cxVeVP4pMVP4RRpV^,pVfVPRRV13,pVf\VR4'd4VPRRV13,pVP V^,V4pTpM;\ P !V\ PVPPR7pVfVPV4pVPV4p WI,p VPV4p W, p VPV 4p VPV 4p V #)NrLrrr])rMrhasattrrrrrr:rdrVrrCr@rrF) rHrrrr input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expandedrCrr@s &&&&& r#rTextEmbeddings.forwards,  #..*K',,.s3K ^  ,,Q  ^ J": "66|D) ^^J/ \\*- r&)rrFr@rrCrr)NNNN r(r)r*r+r,r6rr.r/rrs@@r#r7r7sQ s  r&r7c<aa]tRtRtoRtV3RltRtRtVtV;t #)r=iz Image to Patch Embedding. c <\SV`4VPVPr2VPVP rT\ V\PP4'dTMW"3p\ V\PP4'dTMW33pV^,V^,,V^,V^,,,pW nW0nW@nW`n \P!WEW3R7Vn R#)r4) kernel_sizestrideN)r5r6rfrgrr;rs collectionsabcIterabler?rConv2dr^)rHrGrfrgrr;r?rIs&& r#r6ViltPatchEmbeddings.__init__s !'!2!2F4E4EJ$*$7$79K9Kk#-j+//:R:R#S#SZZdYq #-j+//:R:R#S#SZZdYq !!} 1 5*Q-:VW=:XY $$(&))L:ir&cVPwr#rEW0P8wd \R4hVPPP pVPVP VR74pV#)zeMake sure that the channel dimension of the pixel values match with the one set in the configuration.r)r`r ValueErrorr^r_rrq)rHr}rrrr target_dtypers&& r#rViltPatchEmbeddings.forward$sc2>2D2D/ & ,, ,w --33 OOLOO,O? @r&)rfrr?rgr^rrs@@r#r=r=s jr&r=c<aa]tRtRtoV3RltRRltRtVtV;t#)ViltSelfAttentioni/cF<\SV`4VPVP,^8wd9\ VR4'g'\ RVP RVP R24hVPVn\ VPVP, 4VnVPVP,Vn\P!VPVPVPR7Vn \P!VPVPVPR7Vn \P!VPVPVPR7Vn\P!VP 4VnR#)rZembedding_sizezThe hidden size z4 is not a multiple of the number of attention heads .)biasN)r5r6r;num_attention_headsrrrtattention_head_size all_head_sizerLinearqkv_biasquerykeyvaluerDattention_probs_dropout_probrFrs&&r#r6ViltSelfAttention.__init__0s1     : : :a ?PVXhHiHi"6#5#5"67334A7  $*#=#= #&v'9'9FHHAN jj/44\BLLQPQR !<< 5H5HR5PQ+dii8P8P.QQ  %/@ **,-=>,,7 _B %--aAq9DDF "/"4"4"6s";t?Q?Q>S"S%**,CD 6G=2O\M]r&)rrrFrrrrNF r(r)r*r+r6rr.r/rrs@@r#rr/sG$r&rcTaa]tRtRtoRtV3RlV3RlltV3RlRltRtVtV;t #)ViltSelfOutputicz The residual connection is defined in ViltLayer instead of here (as is the case with other models), due to the layernorm applied before each block. c <V^8dQhRS[/#rrGr)r!r"s"r#r$ViltSelfOutput.__annotate__i>>z>r&c<\SV`4\P!VPVP4Vn\P !VP4VnR#N) r5r6rrr;denserDrErFrs&&r#r6ViltSelfOutput.__init__isB YYv1163E3EF zz&"<"<= r&ch<V^8dQhRS[PRS[PRS[P/#rr input_tensorreturnrTensor)r!r"s"r#r$r ns1U\\RWR^R^r&cJVPV4pVPV4pV#r r rFrHrrs&&&r#rViltSelfOutput.forwardns$ =1  ]3 r&rrrs@@r#rrcs# >> r&rc<aa]tRtRtoV3RltRRltRtVtV;t#) ViltAttentionitcd<\SV`4\V4Vn\ V4VnR#r )r5r6r attentionroutputrs&&r#r6ViltAttention.__init__us& *62$V, r&c~VPWV4pVPV^,V4pV3VR,,pV#)rZr[rr)rHrrr self_outputsattention_outputrs&&&& r#rViltAttention.forwardzs?~~mEVW ;;|A F#% R(88r&r rrrs@@r#rrts- r&rcPaa]tRtRtoV3RlV3RlltV3RlRltRtVtV;t#)ViltIntermediateic <V^8dQhRS[/#rr)r!r"s"r#r$ViltIntermediate.__annotate__s99z9r&c"<\SV`4\P!VPVP 4Vn\VP\4'd\VP,Vn R#VPVn R#r ) r5r6rrr;intermediate_sizer rs hidden_actstrrintermediate_act_fnrs&&r#r6ViltIntermediate.__init__s` YYv1163K3KL f'' - -'-f.?.?'@D $'-'8'8D $r&cN<V^8dQhRS[PRS[P/#)rrrr)r!r"s"r#r$r's#U\\ellr&cJVPV4pVPV4pV#r r r,rHrs&&r#rViltIntermediate.forwards& =1 00? r&r0rrs@@r#r%r%s99r&r%cPaa]tRtRtoV3RlV3RlltV3RlRltRtVtV;t#) ViltOutputic <V^8dQhRS[/#rr)r!r"s"r#r$ViltOutput.__annotate__r r&c<\SV`4\P!VPVP 4Vn\P!VP4Vn R#r ) r5r6rrr)r;r rDrErFrs&&r#r6ViltOutput.__init__sB YYv779K9KL zz&"<"<= r&ch<V^8dQhRS[PRS[PRS[P/#rr)r!r"s"r#r$r6s1U\\RWR^R^r&cZVPV4pVPV4pW,pV#r rrs&&&r#rViltOutput.forwards, =1  ]3 %4 r&rrrs@@r#r4r4s>> r&r4c@aa]tRtRtoRtV3RltRRltRtVtV;t #) ViltLayeriz?This corresponds to the Block class in the timm implementation.cx<\SV`4VPVn^Vn\ V4Vn\ V4Vn\V4Vn \P!VPVPR7Vn\P!VPVPR7VnR#)r4rN)r5r6chunk_size_feed_forward seq_len_dimrrr% intermediater4rrrr;rlayernorm_beforelayernorm_afterrs&&r#r6ViltLayer.__init__s '-'E'E$&v.,V4 ( " V-?-?VEZEZ [!||F,>,>FDYDYZr&c,VPVPV4VVR7pV^,pVR,pWQPVP4,pVP V4pVP V4pVP Wq4pV3V,pV#))rr[)rrBrqrVrCrAr)rHrrrself_attention_outputsr"r layer_outputs&&&& r#rViltLayer.forwards!%  ! !- 0 /"0"  2!4(,)+;+;&ViltEncoder.forward..sm$[q$[s )last_hidden_staterrr')ryrOr r ) rHrrroutput_hidden_states return_dictall_hidden_statesall_self_attentionsr layer_module layer_outputss &&&&&& r#rViltEncoder.forwards#7BD$5b4(4OA#$58H$H!(HYZM)!,M  &91=M?/r&r`cbaa]tRtRtoRV3RlltRtRt]R V3RlRll4tRt Vt V;t #) ViltModelic (<\SV`V4Wn\V4Vn\ V4Vn\P!VPVPR7Vn V'd \V4MRVn VP4R#)z^ add_pooling_layer (bool, *optional*, defaults to `True`): Whether to add a pooling layer rN)r5r6rGr1rrJencoderrrr;r layernorm ViltPoolerpooler post_init)rHrGadd_pooling_layerrIs&&&r#r6ViltModel.__init__si   (0"6* f&8&8f>S>ST,=j(4  r&cBVPPP#r rr8rrHs&r#get_input_embeddingsViltModel.get_input_embeddingss..>>>r&c:WPPnR#r r|)rHrs&&r#set_input_embeddingsViltModel.set_input_embeddingss:?''7r&c<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[R,R S[R,R S[R,R S[R,R S[S[S[P,,/ #)rrNrrr}r~rrrrrXrYr)r LongTensorrrtboolr r )r!r"s"r#r$ViltModel.__annotate__sp p ##d*p ))D0p ((4/ p ''$. p $$t+ p ((4/p ''$.p "Djp  $;p #Tkp D[p  $eE,=,=&> >p r&c V eT MVPPp V eT MVPPp V eT MVPPp VeVe \ R4hVe#VP W4VP 4p M#VeVP 4RR p M \ R4hV wrVe VPM VPpVf\P!W3VR7pVeVe \ R4hVfVf \ R4hVeVP^,MVP^,pVV8wd \ R4hVfD\P!VVPPVPP3VR7pVPVVVVVVVVR7wpp\VPVVR 7pVPVVV V V R 7pV^,pVPV4pVP eVP!V4MRpV 'gVV3VR ,,#\#VVVP$VP&R 7#)a  image_embeds (`torch.FloatTensor` of shape `(batch_size, num_patches, hidden_size)`, *optional*): Optionally, instead of passing `pixel_values`, you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert `pixel_values` into patch embeddings. image_token_type_idx (`int`, *optional*): - The token type ids for images. Examples: ```python >>> from transformers import ViltProcessor, ViltModel >>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> # prepare image and text >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> text = "hello world" >>> processor = ViltProcessor.from_pretrained("dandelin/vilt-b32-mlm") >>> model = ViltModel.from_pretrained("dandelin/vilt-b32-mlm") >>> inputs = processor(image, text, return_tensors="pt") >>> outputs = model(**inputs) >>> last_hidden_states = outputs.last_hidden_state ```NzDYou cannot specify both input_ids and inputs_embeds at the same timez5You have to specify either input_ids or inputs_embedsrUzFYou cannot specify both pixel_values and image_embeds at the same timez7You have to specify either pixel_values or image_embedszAThe text inputs and image inputs need to have the same batch size)r)rGrr)rrrXrYr[)rW pooler_outputrrr])rGrrXrYr%warn_if_padding_and_no_attention_maskrMrVrr{r`rfrrrtrurwr rr)rHrrrr}r~rrrrrXrYkwargsrtext_batch_sizerrVimage_batch_sizeembedding_outputencoder_outputssequence_output pooled_outputs&&&&&&&&&&&&, r#rViltModel.forwardsdX2C1N-TXT_T_TqTq$8$D $++JjJj &1%cd d  "  6 6y Q#..*K  &',,.s3KTU U&1#%.%:!! @T@T  !"ZZ/)FPVWN  # (@ef f  !l&:VW W4@4L<--a0R^RdRdefRg  .`a a  %5t{{7M7Mt{{OeOe$fouvJ+/??       !5,; , (.3;;*)  ,, )/!5# ' *!,..98< 8O O4UY #]3ob6II I)-')77&11   r&)rGrrtrurw)T NNNNNNNNNNN) r(r)r*r+r6r~rrrr.r/rrs@@r#rrrrs7"?@p p p p r&rrc8aa]tRtRtoV3RltRtRtVtV;t#)rvic<\SV`4\P!VPVP4Vn\P !4VnR#r )r5r6rrr;r Tanh activationrs&&r#r6ViltPooler.__init__s9 YYv1163E3EF '')r&c\VR,pVPV4pVPV4pV#)rLrY)r r)rHrfirst_token_tensorrs&& r#rViltPooler.forwards2+40 #56  6 r&)rr rrs@@r#rvrvs$ r&rvzU ViLT Model with a language modeling head on top as done during pretraining. cfaa]tRtRtoRR/tV3RltRtRt]R V3RlRll4t R t Vt V;t #) ViltForMaskedLMizmlm_score.decoder.weightz6vilt.embeddings.text_embeddings.word_embeddings.weightc<\SV`V4\V4Vn\ V4VnVP 4R#r )r5r6rrra ViltMLMHead mlm_scorerxrs&&r#r6ViltForMaskedLM.__init__s4  f% $V, r&c.VPP#r )rdecoderr}s&r#get_output_embeddings%ViltForMaskedLM.get_output_embeddingss~~%%%r&c\WPnVPVPnR#r )rrr)rHnew_embeddingss&&r#set_output_embeddings%ViltForMaskedLM.set_output_embeddingss!/,11r&c<V^8dQhRS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,RS[PR,R S[PR,R S[R,R S[R,R S[R,R S[S[S[P,,/ #rrNrrr}r~rrlabelsrrXrYr)rrrrr r )r!r"s"r#r$ViltForMaskedLM.__annotate__sp p ##d*p ))D0p ((4/ p ''$. p $$t+ p ((4/p ''$.p   4'p  $;p #Tkp D[p  % 1 12 2p r&c V eT MVPPp VPVVVVVVVV V V R7 p V R,wrVeVP^,MVP^,pVRRV13,VRVR13,ppVP V4pRpVeb\ 4pVP VP4pV!VPRVPP4VPR44pV 'g"V3V R,,pVe V3V,#T#\VVV PV PR7#)a image_embeds (`torch.FloatTensor` of shape `(batch_size, num_patches, hidden_size)`, *optional*): Optionally, instead of passing `pixel_values`, you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert `pixel_values` into patch embeddings. labels (*torch.LongTensor* 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]* Examples: ```python >>> from transformers import ViltProcessor, ViltForMaskedLM >>> import httpx >>> from io import BytesIO >>> from PIL import Image >>> import re >>> import torch >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> text = "a bunch of [MASK] laying on a [MASK]." >>> processor = ViltProcessor.from_pretrained("dandelin/vilt-b32-mlm") >>> model = ViltForMaskedLM.from_pretrained("dandelin/vilt-b32-mlm") >>> # prepare inputs >>> encoding = processor(image, text, return_tensors="pt") >>> # forward pass >>> outputs = model(**encoding) >>> tl = len(re.findall("\[MASK\]", text)) >>> inferred_token = [text] >>> # gradually fill in the MASK tokens, one by one >>> with torch.no_grad(): ... for i in range(tl): ... encoded = processor.tokenizer(inferred_token) ... input_ids = torch.tensor(encoded.input_ids) ... encoded = encoded["input_ids"][0][1:-1] ... outputs = model(input_ids=input_ids, pixel_values=encoding.pixel_values) ... mlm_logits = outputs.logits[0] # shape (seq_len, vocab_size) ... # only take into account text features (minus CLS and SEP token) ... mlm_logits = mlm_logits[1 : input_ids.shape[1] - 1, :] ... mlm_values, mlm_ids = mlm_logits.softmax(dim=-1).max(dim=-1) ... # only take into account text ... mlm_values[torch.tensor(encoded) != 103] = 0 ... select = mlm_values.argmax().item() ... encoded[select] = mlm_ids[select].item() ... inferred_token = [processor.decode(encoded)] >>> selected_token = "" >>> encoded = processor.tokenizer(inferred_token) >>> output = processor.decode(encoded.input_ids[0], skip_special_tokens=True) >>> print(output) a bunch of cats laying on a couch. ```N rrr}r~rrrrXrY:NrNrLrNNrrrrr]) rGrYrar`rrrqrVrirr rr)rHrrrr}r~rrrrrXrYrrrr text_seq_len text_featuresr mlm_logitsmasked_lm_lossloss_fctrs&&&&&&&&&&&&, r#rViltForMaskedLM.forwardsTV&1%S>STr&clVPV4pVPV4pVPV4pV#r )r rrr1s&&r#r#ViltPredictionHeadTransform.forward2s4 =1 --m< }5 r&)rr rrrs@@r#rr(sUr&rc8aa]tRtRtoV3RltRtRtVtV;t#)ri9c<\SV`4Wn\V4Vn\ P !VPVP4Vn R#r ) r5r6rGr transformrrr;rrrs&&r#r6ViltMLMHead.__init__:s=  4V<yy!3!3V5F5FG r&cJVPV4pVPV4pV#r )rr)rHrs&&r#rViltMLMHead.forward@s" NN1  LLOr&)rGrrrrs@@r#rr9sH r&rz Vilt Model transformer with a classifier head on top (a linear layer on top of the final hidden state of the [CLS] token) for visual question answering, e.g. for VQAv2. cRaa]tRtRtoV3Rlt]RV3RlRll4tRtVtV;t #)ViltForQuestionAnsweringiFc <\SV`V4VPVn\V4Vn\ P !\ P!VPVP^,4\ P!VP^,4\ P!4\ P!VP^,VP44Vn VP4R#)rN) r5r6 num_labelsrrrar Sequentialrr;rGELU classifierrxrs&&r#r6!ViltForQuestionAnswering.__init__Ms   ++f% -- IIf((&*<*>> from transformers import ViltProcessor, ViltForQuestionAnswering >>> import httpx >>> from io import BytesIO >>> from PIL import Image >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> text = "How many cats are there?" >>> processor = ViltProcessor.from_pretrained("dandelin/vilt-b32-finetuned-vqa") >>> model = ViltForQuestionAnswering.from_pretrained("dandelin/vilt-b32-finetuned-vqa") >>> # prepare inputs >>> encoding = processor(image, text, return_tensors="pt") >>> # forward pass >>> outputs = model(**encoding) >>> logits = outputs.logits >>> idx = logits.argmax(-1).item() >>> print("Predicted answer:", model.config.id2label[idx]) Predicted answer: 2 ```Nrrr)rGrYrarrrqrVrrb binary_cross_entropy_with_logitsr`r rr)rHrrrr}r~rrrrrXrYrrrrrrs&&&&&&&&&&&&, r#r ViltForQuestionAnswering.forward^sf&1%