o wi@sdZddlZddlmZddlmZmZddlZddlZddlm Z ddl m Z m Z m Z ddlmZmZdd lmZdd lmZmZdd lmZdd lmZdd lmZmZmZddlmZe e!Z"eeddGdddeZ#eeddGdddeZ$eeddGdddeZ%eeddGdddeZ&eeddGdddeZ'eeddGd d!d!eZ(eed"dGd#d$d$eZ)eed%dGd&d'd'eZ*eed(dGd)d*d*eZ+eed+dGd,d-d-eZ,Gd.d/d/e j-Z.Gd0d1d1e j-Z/Gd2d3d3e j-Z0Gd4d5d5e j-Z1Gd6d7d7e j-Z2Gd8d9d9e j-Z3Gd:d;d;e j-Z4Gdd?d?e j-Z6Gd@dAdAe j-Z7GdBdCdCe j-Z8GdDdEdEe j-Z9eGdFdGdGeZ:edHdGdIdJdJe:Z;dKdLZedRdGdSdTdTe:Z?edUdGdVdWdWe:Z@edXdGdYdZdZe:ZAed[dGd\d]d]e:ZBed^dGd_d`d`e:ZCeGdadbdbe:ZDeGdcdddde:ZEgdeZFdS)fzPyTorch LUKE model.N) dataclass)OptionalUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FNgelu)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPooling)PreTrainedModel)apply_chunking_to_forward) ModelOutputauto_docstringlogging) LukeConfigz3 Base class for outputs of the LUKE model. ) custom_introc@>eZdZUdZdZeejed<dZ ee ejdfed<dS)BaseLukeModelOutputWithPoolingax pooler_output (`torch.FloatTensor` of shape `(batch_size, hidden_size)`): Last layer hidden-state of the first token of the sequence (classification token) further processed by a Linear layer and a Tanh activation function. entity_last_hidden_state (`torch.FloatTensor` of shape `(batch_size, entity_length, hidden_size)`): Sequence of entity hidden-states at the output of the last layer of the model. entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nentity_last_hidden_state.entity_hidden_states __name__ __module__ __qualname____doc__rrtorch FloatTensor__annotations__rtupler$r$c/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/transformers/models/luke/modeling_luke.pyr&s  rzV Base class for model's outputs, with potential hidden states and attentions. c@r)BaseLukeModelOutputa entity_last_hidden_state (`torch.FloatTensor` of shape `(batch_size, entity_length, hidden_size)`): Sequence of entity hidden-states at the output of the last layer of the model. entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr.rrr$r$r$r%r&=s  r&c@seZdZUdZdZeejed<dZ eejed<dZ eejed<dZ eejed<dZ eejed<dZ eeejed<dZeeejd fed <dZeeejd fed <dS) LukeMaskedLMOutputa: loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): The sum of masked language modeling (MLM) loss and entity prediction loss. mlm_loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Masked language modeling (MLM) loss. mep_loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Masked entity prediction (MEP) loss. logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`): Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). entity_logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`): Prediction scores of the entity prediction head (scores for each entity vocabulary token before SoftMax). entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nlossmlm_lossmep_losslogits entity_logits hidden_states.r attentions)rrrrr(rr r!r"r)r*r+r,r-r#rr.r$r$r$r%r'Qs r'z2 Outputs of entity classification models. c@eZdZUdZdZeejed<dZ eejed<dZ ee ejdfed<dZ ee ejdfed<dZ ee ejdfed<dS) EntityClassificationOutput loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Classification loss. logits (`torch.FloatTensor` of shape `(batch_size, config.num_labels)`): Classification scores (before SoftMax). entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr(r+.r-rr.rrrrr(rr r!r"r+r-r#rr.r$r$r$r%r0s  r0z7 Outputs of entity pair classification models. c@r/) EntityPairClassificationOutputr1Nr(r+.r-rr.r2r$r$r$r%r4r3r4z7 Outputs of entity span classification models. c@r/) EntitySpanClassificationOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Classification loss. logits (`torch.FloatTensor` of shape `(batch_size, entity_length, config.num_labels)`): Classification scores (before SoftMax). entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr(r+.r-rr.r2r$r$r$r%r5r3r5z4 Outputs of sentence classification models. c@r/) LukeSequenceClassifierOutputa 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). entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr(r+.r-rr.r2r$r$r$r%r6r3r6z@ Base class for outputs of token classification models. c@r/) LukeTokenClassifierOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Classification loss. logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.num_labels)`): Classification scores (before SoftMax). entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr(r+.r-rr.r2r$r$r$r%r7r3r7z/ Outputs of question answering models. c@seZdZUdZdZeejed<dZ eejed<dZ eejed<dZ ee ejdfed<dZ ee ejdfed<dZee ejdfed <dS) LukeQuestionAnsweringModelOutputa loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Total span extraction loss is the sum of a Cross-Entropy for the start and end positions. entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr( start_logits end_logits.r-rr.)rrrrr(rr r!r"r9r:r-r#rr.r$r$r$r%r8s  r8z, Outputs of multiple choice models. c@r/) LukeMultipleChoiceModelOutputa loss (`torch.FloatTensor` of shape *(1,)*, *optional*, returned when `labels` is provided): Classification loss. logits (`torch.FloatTensor` of shape `(batch_size, num_choices)`): *num_choices* is the second dimension of the input tensors. (see *input_ids* above). Classification scores (before SoftMax). entity_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of shape `(batch_size, entity_length, hidden_size)`. Entity hidden-states of the model at the output of each layer plus the initial entity embedding outputs. Nr(r+.r-rr.r2r$r$r$r%r;s  r;cs:eZdZdZfddZ    d ddZddZZS) LukeEmbeddingszV Same as BertEmbeddings with a tiny tweak for positional embeddings indexing. csttj|j|j|jd|_t|j|j|_ t|j |j|_ tj |j|j d|_ t|j|_|j|_tj|j|j|jd|_ dS)N padding_idxeps)super__init__r Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingstype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutr>selfconfig __class__r$r%rB(s   zLukeEmbeddings.__init__Nc Cs|dur|durt||j|j}n||}|dur!|}n|dd}|dur8tj|tj|j jd}|durA| |}| |}| |}|||}| |}||}|S)Ndtypedevice)"create_position_ids_from_input_idsr>torY&create_position_ids_from_inputs_embedssizer zeroslong position_idsrGrIrKrLrP) rR input_idstoken_type_idsr` inputs_embeds input_shaperIrK embeddingsr$r$r%forward9s"        zLukeEmbeddings.forwardcCsN|dd}|d}tj|jd||jdtj|jd}|d|S)z We are provided embeddings directly. We cannot infer which are padded so just generate sequential position ids. Args: inputs_embeds: torch.Tensor Returns: torch.Tensor NrVrrWr)r]r aranger>r_rY unsqueezeexpand)rRrcrdsequence_lengthr`r$r$r%r\Zs  z5LukeEmbeddings.create_position_ids_from_inputs_embeds)NNNN)rrrrrBrfr\ __classcell__r$r$rTr%r<#s  !r<csFeZdZdeffdd Z d dejdejdeejfdd ZZ S) LukeEntityEmbeddingsrScst||_tj|j|jdd|_|j|jkr$tj |j|jdd|_ t|j |j|_ t|j |j|_tj|j|jd|_t|j|_dS)Nrr=Fbiasr?)rArBrSrrCentity_vocab_sizeentity_emb_sizeentity_embeddingsrELinearentity_embedding_denserHrIrJrKrLrMrNrOrPrQrTr$r%rBms  zLukeEntityEmbeddings.__init__N entity_idsr`rbc Cs|dur t|}||}|jj|jjkr||}||jdd}|dk | d}||}tj |dd}||j ddjdd}| |}|||}| |}||}|S)Nr)minrVdimgHz>)r zeros_likerqrSrprErsrIclamptype_asrhsumrKrLrP) rRrtr`rbrqrIposition_embedding_maskrKrer$r$r%rf{s       zLukeEntityEmbeddings.forwardN) rrrrrBr LongTensorrrfrkr$r$rTr%rllsrlc4eZdZfddZddZ   d ddZZS) LukeSelfAttentioncst|j|jdkrt|dstd|jd|jd|j|_t|j|j|_|j|j|_|j |_ t |j|j|_ t |j|j|_ t |j|j|_|j rpt |j|j|_t |j|j|_t |j|j|_t |j|_dS)Nrembedding_sizezThe hidden size z4 is not a multiple of the number of attention heads .)rArBrEnum_attention_headshasattr ValueErrorintattention_head_size all_head_sizeuse_entity_aware_attentionrrrquerykeyvalue w2e_query e2w_query e2e_queryrNattention_probs_dropout_probrPrQrTr$r%rBs&  zLukeSelfAttention.__init__cCs6|dd|j|jf}|j|}|ddddS)NrVrrr )r]rrviewpermute)rRx new_x_shaper$r$r%transpose_for_scoress z&LukeSelfAttention.transpose_for_scoresNFc Cs|d}|dur |}n tj||gdd}|||}|||} |jr|dur|||} |||} || |} || |} |ddddd|ddf}|ddddd|ddf}|dddd|dddf}|dddd|dddf}t | | dd}t | | dd}t | | dd}t | | dd}tj||gdd}tj||gdd}tj||gdd}n|||}t || dd}|t |j}|dur||}tjj|dd}||}|dur||}t || }|dddd}|dd|jf}|j|}|ddd|ddf}|dur>d}n |dd|dddf}|rU|||f}|S||f}|S)NrrwrVrvr rr)r]r catrrrrrrrrmatmul transposemathsqrtrr functionalsoftmaxrPr contiguousrr) rRword_hidden_statesrattention_mask head_maskoutput_attentions word_sizeconcat_hidden_states key_layer value_layerw2w_query_layerw2e_query_layere2w_query_layere2e_query_layer w2w_key_layer e2w_key_layer w2e_key_layer e2e_key_layerw2w_attention_scoresw2e_attention_scorese2w_attention_scorese2e_attention_scoresword_attention_scoresentity_attention_scoresattention_scores query_layerattention_probs context_layernew_context_layer_shapeoutput_word_hidden_statesoutput_entity_hidden_statesoutputsr$r$r%rfsX           zLukeSelfAttention.forwardNNF)rrrrBrrfrkr$r$rTr%rs  rc8eZdZfddZdejdejdejfddZZS)LukeSelfOutputcsBtt|j|j|_tj|j|jd|_t|j |_ dSNr?) rArBrrrrEdenserLrMrNrOrPrQrTr$r%rB  zLukeSelfOutput.__init__r- input_tensorreturncC&||}||}|||}|Sr~rrPrLrRr-rr$r$r%rf  zLukeSelfOutput.forwardrrrrBr Tensorrfrkr$r$rTr%r  $rcr) LukeAttentioncs*tt||_t||_t|_dSr~)rArBrrRroutputset pruned_headsrQrTr$r%rBs    zLukeAttention.__init__cCtdNz4LUKE does not support the pruning of attention headsNotImplementedError)rRheadsr$r$r% prune_headszLukeAttention.prune_headsNFcCs|d}||||||}|dur|d}|} ntj|dddd}tj||gdd} ||| } | ddd|ddf} |durGd} n | dd|dddf} | | f|dd} | S)Nrrrrw)r]rRr rr)rRrrrrrr self_outputsconcat_self_outputsrattention_outputword_attention_outputentity_attention_outputrr$r$r%rf!s(  zLukeAttention.forwardr)rrrrBrrfrkr$r$rTr%rs rc2eZdZfddZdejdejfddZZS)LukeIntermediatecsDtt|j|j|_t|jt rt |j|_ dS|j|_ dSr~) rArBrrrrEintermediate_sizer isinstance hidden_actstrr intermediate_act_fnrQrTr$r%rBHs   zLukeIntermediate.__init__r-rcCs||}||}|Sr~)rrrRr-r$r$r%rfPs  zLukeIntermediate.forwardrr$r$rTr%rGs rcr) LukeOutputcsBtt|j|j|_tj|j|jd|_t |j |_ dSr) rArBrrrrrErrLrMrNrOrPrQrTr$r%rBXrzLukeOutput.__init__r-rrcCrr~rrr$r$r%rf^rzLukeOutput.forwardrr$r$rTr%rWrrcs4eZdZfddZ   d ddZddZZS) LukeLayercs:t|j|_d|_t||_t||_t||_ dSNr) rArBchunk_size_feed_forward seq_len_dimr attentionr intermediaterrrQrTr$r%rBfs   zLukeLayer.__init__NFc Cs|d}|j|||||d}|dur|d}n tj|dddd}|dd} t|j|j|j|} | ddd|ddf} |durGd} n | dd|dddf} | | f| } | S)Nr)rrrrw)r]rr rrfeed_forward_chunkrr) rRrrrrrrself_attention_outputsconcat_attention_outputr layer_outputword_layer_outputentity_layer_outputr$r$r%rfns*    zLukeLayer.forwardcCs||}|||}|Sr~)rr)rRrintermediate_outputrr$r$r%rs  zLukeLayer.feed_forward_chunkr)rrrrBrfrrkr$r$rTr%res   %rcs0eZdZfddZ     dddZZS) LukeEncodercs:t|_tfddtjD|_d|_dS)Ncsg|]}tqSr$)r).0_rSr$r% sz(LukeEncoder.__init__..F) rArBrSr ModuleListrangenum_hidden_layerslayergradient_checkpointingrQrTrr%rBs   zLukeEncoder.__init__NFTcCs|rdnd}|r dnd} |rdnd} t|jD]7\} } |r'||f}| |f} |dur/|| nd} | |||| |}|d}|durE|d}|rN| |df} q|r[||f}| |f} |sktdd||| || fDSt||| || dS)Nr$rrrcs|] }|dur|VqdSr~r$rvr$r$r% z&LukeEncoder.forward..)last_hidden_stater-r.rr) enumeraterr#r&)rRrrrrroutput_hidden_states return_dictall_word_hidden_statesall_entity_hidden_statesall_self_attentionsi layer_modulelayer_head_mask layer_outputsr$r$r%rfsP       zLukeEncoder.forward)NNFFTrrrrBrfrkr$r$rTr%rs  rcr) LukePoolercs*tt|j|j|_t|_dSr~)rArBrrrrErTanh activationrQrTr$r%rBs zLukePooler.__init__r-rcCs(|dddf}||}||}|S)Nr)rr)rRr-first_token_tensor pooled_outputr$r$r%rfs  zLukePooler.forwardrr$r$rTr%rs rc$eZdZfddZddZZS)EntityPredictionHeadTransformcsVtt|j|j|_t|jt rt |j|_ n|j|_ tj |j|j d|_ dSr)rArBrrrrErprrrrr transform_act_fnrLrMrQrTr$r%rBs  z&EntityPredictionHeadTransform.__init__cCs"||}||}||}|Sr~)rrrLrr$r$r%rfs   z%EntityPredictionHeadTransform.forwardr r$r$rTr%rs  rcr)EntityPredictionHeadcsHt||_t||_tj|j|jdd|_ t t |j|_ dS)NFrm)rArBrSr transformrrrrprodecoder Parameterr r^rnrQrTr$r%rBs  zEntityPredictionHead.__init__cCs||}|||j}|Sr~)rrrnrr$r$r%rfs zEntityPredictionHead.forwardr r$r$rTr%rs rc@s0eZdZeZdZdZddgZdej fddZ dS) LukePreTrainedModellukeTrrlmodulecCst|tjr |jjjd|jjd|jdur|jj dSdSt|tj rO|j dkr2|jj n |jjjd|jjd|j durM|jj|j  dSdSt|tj rd|jj |jjddSdS)zInitialize the weightsg)meanstdNr?)rrrrweightdatanormal_rSinitializer_rangernzero_rC embedding_dimr>rLfill_)rRrr$r$r% _init_weightss       z!LukePreTrainedModel._init_weightsN) rrrr config_classbase_model_prefixsupports_gradient_checkpointing_no_split_modulesrModuler'r$r$r$r%r s rzt The bare LUKE model transformer outputting raw hidden-states for both word tokens and entities without any c seZdZd$dedeffdd ZddZdd Zd d Zd d Z ddZ e             d%de e jde e jde e jde e jde e jde e jde e jde e jde e jde e jde ede ede edeeeffdd Zd!e jde e jfd"d#ZZS)& LukeModelTrSadd_pooling_layercsNt|||_t||_t||_t||_|rt |nd|_ | dS)zv add_pooling_layer (bool, *optional*, defaults to `True`): Whether to add a pooling layer N) rArBrSr<rerlrqrencoderrpooler post_init)rRrSr.rTr$r%rB,s     zLukeModel.__init__cC|jjSr~rerGrRr$r$r%get_input_embeddings=rzLukeModel.get_input_embeddingscC ||j_dSr~r3rRrr$r$r%set_input_embeddings@ zLukeModel.set_input_embeddingscCs|jjSr~rqr4r$r$r%get_entity_embeddingsCrzLukeModel.get_entity_embeddingscCs ||j_dSr~r:r7r$r$r%set_entity_embeddingsFr9zLukeModel.set_entity_embeddingscCrrr)rRheads_to_pruner$r$r% _prune_headsIrzLukeModel._prune_headsNrarrbr`rtentity_attention_maskentity_token_type_idsentity_position_idsrrcrrrrc Cs| dur| n|jj} | dur| n|jj} | dur| n|jj} |dur*| dur*td|dur9||||}n| durF| dd}ntd|\}}|durU|jn| j}|duretj ||f|d}|durrtj |tj |d}|dur|d}|durtj ||f|d}|durtj ||ftj |d}| | |jj } |j|||| d}|||}|durd}n||||}|j|||| | | | d }|d }|jdur||nd}| s||f|ddSt|||j|j|j|jd S) uz entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. Examples: ```python >>> from transformers import AutoTokenizer, LukeModel >>> tokenizer = AutoTokenizer.from_pretrained("studio-ousia/luke-base") >>> model = LukeModel.from_pretrained("studio-ousia/luke-base") # Compute the contextualized entity representation corresponding to the entity mention "Beyoncé" >>> text = "Beyoncé lives in Los Angeles." >>> entity_spans = [(0, 7)] # character-based entity span corresponding to "Beyoncé" >>> encoding = tokenizer(text, entity_spans=entity_spans, add_prefix_space=True, return_tensors="pt") >>> outputs = model(**encoding) >>> word_last_hidden_state = outputs.last_hidden_state >>> entity_last_hidden_state = outputs.entity_last_hidden_state # Input Wikipedia entities to obtain enriched contextualized representations of word tokens >>> text = "Beyoncé lives in Los Angeles." >>> entities = [ ... "Beyoncé", ... "Los Angeles", ... ] # Wikipedia entity titles corresponding to the entity mentions "Beyoncé" and "Los Angeles" >>> entity_spans = [ ... (0, 7), ... (17, 28), ... ] # character-based entity spans corresponding to "Beyoncé" and "Los Angeles" >>> encoding = tokenizer( ... text, entities=entities, entity_spans=entity_spans, add_prefix_space=True, return_tensors="pt" ... ) >>> outputs = model(**encoding) >>> word_last_hidden_state = outputs.last_hidden_state >>> entity_last_hidden_state = outputs.entity_last_hidden_state ```NzDYou cannot specify both input_ids and inputs_embeds at the same timerVz5You have to specify either input_ids or inputs_embeds)rYrWr)rar`rbrc)rrrrrr)r pooler_outputr-r.rr)rSrruse_return_dictr%warn_if_padding_and_no_attention_maskr]rYr onesr^r_ get_head_maskrreget_extended_attention_maskrqr/r0rr-r.rr)rRrarrbr`rtr?r@rArrcrrrrd batch_size seq_lengthrYentity_seq_lengthword_embedding_outputextended_attention_maskentity_embedding_outputencoder_outputssequence_outputrr$r$r%rfLspI     zLukeModel.forwardword_attention_maskcCs|}|durtj||gdd}|dkr$|dddddddf}n|dkr7|ddddddf}n td|jd|j|jd}d |t|jj}|S) ac Makes broadcastable attention and causal masks so that future and masked tokens are ignored. Arguments: word_attention_mask (`torch.LongTensor`): Attention mask for word tokens with ones indicating tokens to attend to, zeros for tokens to ignore. entity_attention_mask (`torch.LongTensor`, *optional*): Attention mask for entity tokens with ones indicating tokens to attend to, zeros for tokens to ignore. Returns: `torch.Tensor` The extended attention mask, with a the same dtype as `attention_mask.dtype`. NrVrwr rz&Wrong shape for attention_mask (shape ))rXr) r rrxrshaper[rXfinforu)rRrPr?rrLr$r$r%rGs  z%LukeModel.get_extended_attention_mask)T) NNNNNNNNNNNNN)rrrrboolrBr5r8r;r<r>rrr rr!rr#rrfrGrkr$r$rTr%r-&sp       r-cCs2||}tj|dd||}||S)a Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols are ignored. This is modified from fairseq's `utils.make_positions`. Args: x: torch.Tensor x: Returns: torch.Tensor rrw)nerr cumsumr{r_)rar>maskincremental_indicesr$r$r%rZs  rZcs0eZdZdZfddZddZddZZS) LukeLMHeadz*Roberta Head for masked language modeling.csdtt|j|j|_tj|j|jd|_t|j|j |_ t t |j |_|j|j _dSr)rArBrrrrErrLrM layer_normrDrrr r^rnrQrTr$r%rBs zLukeLMHead.__init__cKs*||}t|}||}||}|Sr~)rr rZr)rRfeatureskwargsrr$r$r%rf$s   zLukeLMHead.forwardcCs,|jjjjdkr|j|j_dS|jj|_dS)Nmeta)rrnrYtyper4r$r$r% _tie_weights.szLukeLMHead._tie_weights)rrrrrBrfr_rkr$r$rTr%rYs    rYz The LUKE model with a language modeling head and entity prediction head on top for masked language modeling and masked entity prediction. c$s eZdZgdZfddZfddZddZdd Ze dd e e j d e e j d e e j de e j de e j de e j de e j de e j de e j de e j de e j de e j de e de e de e deeeff ddZZS)LukeForMaskedLM)zlm_head.decoder.weightzlm_head.decoder.biasz!entity_predictions.decoder.weightcs@t|t||_t||_t||_t |_ | dSr~) rArBr-rrYlm_headrentity_predictionsrrloss_fnr1rQrTr$r%rB@s      zLukeForMaskedLM.__init__cs$t||jj|jjjdSr~)rA tie_weights_tie_or_clone_weightsrbrrrqr4rTr$r%rdMs zLukeForMaskedLM.tie_weightscCr2r~rarr4r$r$r%get_output_embeddingsQrz%LukeForMaskedLM.get_output_embeddingscCr6r~rf)rRnew_embeddingsr$r$r%set_output_embeddingsTr9z%LukeForMaskedLM.set_output_embeddingsNrarrbr`rtr?r@rAlabels entity_labelsrrcrrrrcCs.|dur|n|jj}|j||||||||| | | |dd }d}d}||j}| durE| |j} ||d|jj | d}|durE|}d}d}|j durr| |j }| durr||d|jj | d}|durn|}n||}|st dd||||||j|j|jfDSt||||||j|j|jdS)aC entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. 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]` entity_labels (`torch.LongTensor` of shape `(batch_size, entity_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]` NT rarrbr`rtr?r@rArrcrrrrVcsrr~r$rr$r$r%rs  z*LukeForMaskedLM.forward..)r(r)r*r+r,r-rr.)rSrCrrarr[rYrcrrDrrbror#r-rr.r')rRrarrbr`rtr?r@rArjrkrrcrrrrr(r)r+r*r,r$r$r%rfWsn1    zLukeForMaskedLM.forwardNNNNNNNNNNNNNNN)rrr_tied_weights_keysrBrdrgrirrr rr!rTrr#r'rfrkr$r$rTr%r`7sn        r`z The LUKE model with a classification head on top (a linear layer on top of the hidden state of the first entity token) for entity classification tasks, such as Open Entity. c"eZdZfddZe              ddeejdeejdeejdeejdeejd eejd eejd eejd eejd eejdeejdee dee dee de e e ffddZ ZS)LukeForEntityClassificationcsJt|t||_|j|_t|j|_t |j |j|_ | dSr~ rArBr-r num_labelsrrNrOrPrrrE classifierr1rQrTr$r%rBs   z$LukeForEntityClassification.__init__Nrarrbr`rtr?r@rArrcrjrrrrcCs|dur|n|jj}|j||||||||| | | | dd }|jdddddf}||}||}d}| dur[| |j} | jdkrKt j || }nt j | d| d|}|sntdd|||j|j|jfDSt|||j|j|jd S) u entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. labels (`torch.LongTensor` of shape `(batch_size,)` or `(batch_size, num_labels)`, *optional*): Labels for computing the classification loss. If the shape is `(batch_size,)`, the cross entropy loss is used for the single-label classification. In this case, labels should contain the indices that should be in `[0, ..., config.num_labels - 1]`. If the shape is `(batch_size, num_labels)`, the binary cross entropy loss is used for the multi-label classification. In this case, labels should only contain `[0, 1]`, where 0 and 1 indicate false and true, respectively. Examples: ```python >>> from transformers import AutoTokenizer, LukeForEntityClassification >>> tokenizer = AutoTokenizer.from_pretrained("studio-ousia/luke-large-finetuned-open-entity") >>> model = LukeForEntityClassification.from_pretrained("studio-ousia/luke-large-finetuned-open-entity") >>> text = "Beyoncé lives in Los Angeles." >>> entity_spans = [(0, 7)] # character-based entity span corresponding to "Beyoncé" >>> inputs = tokenizer(text, entity_spans=entity_spans, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits >>> predicted_class_idx = logits.argmax(-1).item() >>> print("Predicted class:", model.config.id2label[predicted_class_idx]) Predicted class: person ```NTrlrrrVcsrr~r$rr$r$r%r?z6LukeForEntityClassification.forward..r(r+r-rr.)rSrCrrrPrsr[rYndimrr cross_entropy binary_cross_entropy_with_logitsrr{r#r-rr.r0rRrarrbr`rtr?r@rArrcrjrrrrfeature_vectorr+r(r$r$r%rfsH>     z#LukeForEntityClassification.forwardNNNNNNNNNNNNNN)rrrrBrrr rr!rTrr#r0rfrkr$r$rTr%rp`        rpz The LUKE model with a classification head on top (a linear layer on top of the hidden states of the two entity tokens) for entity pair classification tasks, such as TACRED. c"seZdZfddZe              ddeejdeejdeejdeejdeejd eejd eejd eejd eejd eejdeejdee dee dee de e e ffddZ ZS)LukeForEntityPairClassificationcsPt|t||_|j|_t|j|_t |j d|jd|_ | dS)NrFrqrQrTr$r%rBUs   z(LukeForEntityPairClassification.__init__Nrarrbr`rtr?r@rArrcrjrrrrcCs|dur|n|jj}|j||||||||| | | | dd }tj|jdddddf|jdddddfgdd}||}||}d}| durl| |j } | j dkr\t j || }nt j |d| d|}|stdd |||j|j|jfDSt|||j|j|jd S) u entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. labels (`torch.LongTensor` of shape `(batch_size,)` or `(batch_size, num_labels)`, *optional*): Labels for computing the classification loss. If the shape is `(batch_size,)`, the cross entropy loss is used for the single-label classification. In this case, labels should contain the indices that should be in `[0, ..., config.num_labels - 1]`. If the shape is `(batch_size, num_labels)`, the binary cross entropy loss is used for the multi-label classification. In this case, labels should only contain `[0, 1]`, where 0 and 1 indicate false and true, respectively. Examples: ```python >>> from transformers import AutoTokenizer, LukeForEntityPairClassification >>> tokenizer = AutoTokenizer.from_pretrained("studio-ousia/luke-large-finetuned-tacred") >>> model = LukeForEntityPairClassification.from_pretrained("studio-ousia/luke-large-finetuned-tacred") >>> text = "Beyoncé lives in Los Angeles." >>> entity_spans = [ ... (0, 7), ... (17, 28), ... ] # character-based entity spans corresponding to "Beyoncé" and "Los Angeles" >>> inputs = tokenizer(text, entity_spans=entity_spans, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits >>> predicted_class_idx = logits.argmax(-1).item() >>> print("Predicted class:", model.config.id2label[predicted_class_idx]) Predicted class: per:cities_of_residence ```NTrlrrrwrVcsrr~r$rr$r$r%rrtz:LukeForEntityPairClassification.forward..ru)rSrCrr rrrPrsr[rYrvrrrwrxrr{r#r-rr.r4ryr$r$r%rfasLA0     z'LukeForEntityPairClassification.forwardr{)rrrrBrrr rr!rTrr#r4rfrkr$r$rTr%r}Nr|r}z The LUKE model with a span classification head on top (a linear layer on top of the hidden states output) for tasks such as named entity recognition. c&seZdZfddZe                ddeejdeejdeejdeejdeejd eejd eejd eejd eejd eejdeejdeejdeejdee dee dee de e e ff"ddZ ZS)LukeForEntitySpanClassificationcsNt|t||_|j|_t|j|_t |j d|j|_ | dS)Nr rqrQrTr$r%rBs   z(LukeForEntitySpanClassification.__init__Nrarrbr`rtr?r@rAentity_start_positionsentity_end_positionsrrcrjrrrrcCs|dur|n|jj}|j||||||||| | ||dd }|jd}| ddd|} | j|jjkr:| |jj} t |jd| }| ddd|} | j|jjkrZ| |jj} t |jd| }t j |||j gdd}| |}||}d}| dur| |j} | jdkrtj|d|j| d}ntj|d| d|}|stdd |||j|j|jfDSt|||j|j|jd S) u entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. entity_start_positions (`torch.LongTensor`): The start positions of entities in the word token sequence. entity_end_positions (`torch.LongTensor`): The end positions of entities in the word token sequence. labels (`torch.LongTensor` of shape `(batch_size, entity_length)` or `(batch_size, entity_length, num_labels)`, *optional*): Labels for computing the classification loss. If the shape is `(batch_size, entity_length)`, the cross entropy loss is used for the single-label classification. In this case, labels should contain the indices that should be in `[0, ..., config.num_labels - 1]`. If the shape is `(batch_size, entity_length, num_labels)`, the binary cross entropy loss is used for the multi-label classification. In this case, labels should only contain `[0, 1]`, where 0 and 1 indicate false and true, respectively. Examples: ```python >>> from transformers import AutoTokenizer, LukeForEntitySpanClassification >>> tokenizer = AutoTokenizer.from_pretrained("studio-ousia/luke-large-finetuned-conll-2003") >>> model = LukeForEntitySpanClassification.from_pretrained("studio-ousia/luke-large-finetuned-conll-2003") >>> text = "Beyoncé lives in Los Angeles" # List all possible entity spans in the text >>> word_start_positions = [0, 8, 14, 17, 21] # character-based start positions of word tokens >>> word_end_positions = [7, 13, 16, 20, 28] # character-based end positions of word tokens >>> entity_spans = [] >>> for i, start_pos in enumerate(word_start_positions): ... for end_pos in word_end_positions[i:]: ... entity_spans.append((start_pos, end_pos)) >>> inputs = tokenizer(text, entity_spans=entity_spans, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits >>> predicted_class_indices = logits.argmax(-1).squeeze().tolist() >>> for span, predicted_class_idx in zip(entity_spans, predicted_class_indices): ... if predicted_class_idx != 0: ... print(text[span[0] : span[1]], model.config.id2label[predicted_class_idx]) Beyoncé PER Los Angeles LOC ```NTrlrVrvrrwcsrr~r$rr$r$r%rertz:LukeForEntitySpanClassification.forward..ru)rSrCrrr]rhrirYr[r gatherrrrPrsrvrrrwrrrrxr{r#r-rr.r5)rRrarrbr`rtr?r@rArrrrcrjrrrrrE start_states end_statesrzr+r(r$r$r%rfsZO       z'LukeForEntitySpanClassification.forward)NNNNNNNNNNNNNNNN)rrrrBrrr rr!rTrr#r5rfrkr$r$rTr%r~sl        r~z The LUKE Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks. c"ro)LukeForSequenceClassificationcsZt||j|_t||_t|jdur|jn|j|_ t |j |j|_ | dSr~rArBrrr-rrrNclassifier_dropoutrOrPrrrErsr1rQrTr$r%rB{s   z&LukeForSequenceClassification.__init__Nrarrbr`rtr?r@rArrcrjrrrrcCs|dur|n|jj}|j||||||||| | | | dd }|j}||}||}d}| dur| |j} |jjdur_|j dkrEd|j_n|j dkr[| j t j ksV| j t j kr[d|j_nd|j_|jjdkr}t}|j dkrw||| }n+||| }n%|jjdkrt}||d|j | d}n|jjdkrt}||| }|std d |||j|j|jfDSt|||j|j|jd S) a entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. labels (`torch.LongTensor` 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). NTrlr regressionsingle_label_classificationmulti_label_classificationrVcsrr~r$rr$r$r%rrtz8LukeForSequenceClassification.forward..ru)rSrCrrBrPrsr[rY problem_typerrrXr r_rrsqueezerrrr#r-rr.r6)rRrarrbr`rtr?r@rArrcrjrrrrrr+r(loss_fctr$r$r%rfsd+      "       z%LukeForSequenceClassification.forwardr{)rrrrBrrr rr!rTrr#r6rfrkr$r$rTr%rtr|rz The LUKE Model with a token classification head on top (a linear layer on top of the hidden-states output). To solve Named-Entity Recognition (NER) task using LUKE, `LukeForEntitySpanClassification` is more suitable than this class. c"ro)LukeForTokenClassificationcs^t||j|_t|dd|_t|jdur|jn|j|_ t |j |j|_ | dSNF)r.rrQrTr$r%rBs  z#LukeForTokenClassification.__init__Nrarrbr`rtr?r@rArrcrjrrrrcCs|dur|n|jj}|j||||||||| | | | dd }|j}||}||}d}| durE| |j} t}|| d|j | d}|sXt dd|||j |j |jfDSt|||j |j |jdS)aM entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See `input_ids` above) NTrlrVcsrr~r$rr$r$r%rQrtz5LukeForTokenClassification.forward..ru)rSrCrrrPrsr[rYrrrrr#r-rr.r7)rRrarrbr`rtr?r@rArrcrjrrrrrOr+r(rr$r$r%rfsF+   z"LukeForTokenClassification.forwardr{)rrrrBrrr rr!rTrr#r7rfrkr$r$rTr%rs`        rc$seZdZfddZe               ddeejdeejdeejdeejdeejd eejd eejd eejd eejd eejdeejdeejdee dee dee de e e ff ddZ ZS)LukeForQuestionAnsweringcs@t||j|_t|dd|_t|j|j|_| dSr) rArBrrr-rrrrrE qa_outputsr1rQrTr$r%rBbs  z!LukeForQuestionAnswering.__init__Nrarrbr`rtr?r@rArrcstart_positions end_positionsrrrrcCsD|dur|n|jj}|j||||||||| | | |dd }|j}||}|jddd\}}|d}|d}d}| dur| durt| dkrN| d} t| dkr[| d} |d}| d|| d|t |d}||| }||| }||d }|st d d ||||j |j |jfDSt||||j |j |jd S) a  entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. NTrlrrVrwr) ignore_indexrcsrr~r$rr$r$r%rs z3LukeForQuestionAnswering.forward..)r(r9r:r-rr.)rSrCrrrsplitrlenr]clamp_rr#r-rr.r8)rRrarrbr`rtr?r@rArrcrrrrrrrOr+r9r: total_loss ignored_indexr start_lossend_lossr$r$r%rfmsh(             z LukeForQuestionAnswering.forwardrm)rrrrBrrr rr!rTrr#r8rfrkr$r$rTr%r`sf        rc"ro)LukeForMultipleChoicecsPt|t||_t|jdur|jn|j|_t |j d|_ | dSr) rArBr-rrrNrrOrPrrrErsr1rQrTr$r%rBs   zLukeForMultipleChoice.__init__Nrarrbr`rtr?r@rArrcrjrrrrcCs|dur|n|jj}|dur|jdn| jd}|dur%|d|dnd}|dur4|d|dnd}|durC|d|dnd}|durR|d|dnd}| dure| d| d| dnd} |durt|d|dnd}|dur|d|dnd}|dur|d|dnd}|dur|d|d|dnd}|j||||||||| | | | dd }|j}||}||}|d|}d}| dur| |j } t }||| }|st dd|||j |j|jfDSt|||j |j|jd S) a^ input_ids (`torch.LongTensor` of shape `(batch_size, num_choices, sequence_length)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) token_type_ids (`torch.LongTensor` of shape `(batch_size, num_choices, sequence_length)`, *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 (`torch.LongTensor` of shape `(batch_size, num_choices, sequence_length)`, *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) entity_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`): Indices of entity tokens in the entity vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. entity_attention_mask (`torch.FloatTensor` of shape `(batch_size, entity_length)`, *optional*): Mask to avoid performing attention on padding entity token indices. Mask values selected in `[0, 1]`: - 1 for entity tokens that are **not masked**, - 0 for entity tokens that are **masked**. entity_token_type_ids (`torch.LongTensor` of shape `(batch_size, entity_length)`, *optional*): Segment token indices to indicate first and second portions of the entity token inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *portion A* entity token, - 1 corresponds to a *portion B* entity token. entity_position_ids (`torch.LongTensor` of shape `(batch_size, entity_length, max_mention_length)`, *optional*): Indices of positions of each input entity in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. inputs_embeds (`torch.FloatTensor` of shape `(batch_size, num_choices, sequence_length, 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. labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See `input_ids` above) NrrVrvTrlcsrr~r$rr$r$r%rdrz0LukeForMultipleChoice.forward..ru)rSrCrRrr]rrBrPrsr[rYrr#r-rr.r;)rRrarrbr`rtr?r@rArrcrjrrr num_choicesrrr+reshaped_logitsr(rr$r$r%rfs~C      zLukeForMultipleChoice.forwardr{)rrrrBrrr rr!rTrr#r;rfrkr$r$rTr%rs`        r) rpr}r~rrrrr`r-r)Grr dataclassesrtypingrrr torch.utils.checkpointrtorch.nnrrr activationsr r modeling_layersr modeling_outputsr rmodeling_utilsr pytorch_utilsrutilsrrrconfiguration_luker get_loggerrloggerrr&r'r0r4r5r6r7r8r;r,r<rlrrrrrrrrrrrr-rZrYr`rpr}r~rrrr__all__r$r$r$r%s       I+r04E]|xgv "