U 4AfH@sdZddlZddlZddlmZddlmZmZmZm Z ddl Z ddl Z ddl m Z ddl mZmZmZddlmZdd lmZmZmZmZmZmZmZmZmZdd lmZdd lm Z m!Z!m"Z"dd l#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z)d dl*m+Z+e(,e-Z.dZ/dZ0Gddde j1Z2Gddde j1Z3Gddde j1Z4de3iZ5Gddde j1Z6Gddde j1Z7Gddde j1Z8Gddde j1Z9Gd d!d!e j1Z:Gd"d#d#e j1Z;Gd$d%d%e j1ZGd*d+d+e j1Z?Gd,d-d-e j1Z@Gd.d/d/eZAeGd0d1d1e$ZBd2ZCd3ZDe&d4eCGd5d6d6eAZEe&d7eCGd8d9d9eAZFe&d:eCGd;d<dd?d?eAZHe&d@eCGdAdBdBeAZIe&dCeCGdDdEdEeAZJe&dFeCGdGdHdHeAZKe&dIeCGdJdKdKeAZLe&dLeCGdMdNdNeAZMdS)OzPyTorch ERNIE model.N) dataclass)ListOptionalTupleUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN) )BaseModelOutputWithPastAndCrossAttentions,BaseModelOutputWithPoolingAndCrossAttentions!CausalLMOutputWithCrossAttentionsMaskedLMOutputMultipleChoiceModelOutputNextSentencePredictorOutputQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)PreTrainedModel)apply_chunking_to_forward find_pruneable_heads_and_indicesprune_linear_layer) ModelOutputadd_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardloggingreplace_return_docstrings) ErnieConfigznghuyong/ernie-1.0-base-zhr!c s\eZdZdZfddZd eejeejeejeejeeje ej dddZ Z S) ErnieEmbeddingszGConstruct the embeddings from word, position and token_type embeddings.csttj|j|j|jd|_t|j|j|_ t|j |j|_ |j |_ |j rft|j |j|_tj|j|jd|_t|j|_t|dd|_|jdt|jddd|jd tj|jtjd dddS) N) padding_idxZepsposition_embedding_typeabsolute position_ids)r F) persistenttoken_type_idsdtype)super__init__r Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingsZtype_vocab_sizetoken_type_embeddings use_task_idZtask_type_vocab_sizetask_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutgetattrr%Zregister_buffertorcharangeexpandzerosr'sizelongselfconfig __class__L/tmp/pip-unpacked-wheel-zw5xktn0/transformers/models/ernie/modeling_ernie.pyr.=s( zErnieEmbeddings.__init__Nr) input_idsr* task_type_idsr' inputs_embedspast_key_values_lengthreturncCs2|dk r|}n|dd}|d}|dkrL|jdd|||f}|dkrt|dr|jddd|f} | |d|} | }ntj|tj|jjd}|dkr| |}| |} || } |j dkr| |} | | 7} |j r|dkrtj|tj|jjd}||}| |7} || } || } | S)Nr(r r*rr,devicer&)rCr'hasattrr*rAr?rBrDrRr3r6r%r5r7r8r9r=)rFrLr*rMr'rNrO input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expandedr6 embeddingsr5r8rJrJrKforwardSs6           zErnieEmbeddings.forward)NNNNNr) __name__ __module__ __qualname____doc__r.rr?Z LongTensor FloatTensorintTensorrY __classcell__rJrJrHrKr":s" r"c seZdZd fdd ZejejdddZd ejeejeejeejeejee e ejee e ejdd d Z Z S) ErnieSelfAttentionNcst|j|jdkr>t|ds>td|jd|jd|j|_t|j|j|_|j|j|_t |j|j|_ t |j|j|_ t |j|j|_ t |j|_|pt|dd|_|jdks|jd kr|j|_t d |jd |j|_|j|_dS) NrZembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ()r%r& relative_keyrelative_key_queryr )r-r.r1num_attention_headsrS ValueErrorr_attention_head_size all_head_sizerLinearquerykeyvaluer;Zattention_probs_dropout_probr=r>r%r4r/distance_embedding is_decoderrFrGr%rHrJrKr.s* zErnieSelfAttention.__init__)xrPcCs6|dd|j|jf}||}|ddddS)Nr(rrfr r )rCrgriviewpermute)rFrrZ new_x_shaperJrJrKtranspose_for_scoress z'ErnieSelfAttention.transpose_for_scoresF hidden_statesattention_mask head_maskencoder_hidden_statesencoder_attention_maskpast_key_valueoutput_attentionsrPcCs||}|dk } | r4|dk r4|d} |d} |}n| r^|||} |||} |}nv|dk r|||} |||} tj|d| gdd} tj|d| gdd} n |||} |||} ||} |dk } |jr| | f}t| | dd}|j dks |j dkr| j d| j d}}| r^tj |dtj |j d dd}ntj|tj |j d dd}tj|tj |j d dd}||}|||jd}|j| jd }|j dkrtd | |}||}n4|j dkrtd | |}td | |}|||}|t|j}|dk r:||}tjj|dd}||}|dk rf||}t|| }|dddd }|dd|jf}||}|r||fn|f}|jr||f}|S)Nrr rfZdimr(rdrerQr+zbhld,lrd->bhlrzbhrd,lrd->bhlrr ) rlrurmrnr?catrpmatmulZ transposer%shapeZtensorrDrRrsr@ror4tor,ZeinsummathsqrtrirZ functionalZsoftmaxr=rt contiguousrCrj)rFrwrxryrzr{r|r}Zmixed_query_layerZis_cross_attentionZ key_layerZ value_layerZ query_layer use_cacheZattention_scoresZ query_lengthZ key_lengthZposition_ids_lZposition_ids_rZdistanceZpositional_embeddingZrelative_position_scoresZrelative_position_scores_queryZrelative_position_scores_keyZattention_probsZ context_layerZnew_context_layer_shapeoutputsrJrJrKrYsp             zErnieSelfAttention.forward)N)NNNNNF) rZr[r\r.r?r`rurr^rboolrYrarJrJrHrKrbs$rbcs4eZdZfddZejejejdddZZS)ErnieSelfOutputcsBtt|j|j|_tj|j|jd|_t|j |_ dSNr$) r-r.rrkr1denser9r:r;r<r=rErHrJrKr.s zErnieSelfOutput.__init__rw input_tensorrPcCs&||}||}|||}|SNrr=r9rFrwrrJrJrKrYs  zErnieSelfOutput.forwardrZr[r\r.r?r`rYrarJrJrHrKrs reagerc sveZdZd fdd ZddZd ejeejeejeejeejee e ejee e ejddd Z Z S) ErnieAttentionNcs4tt|j||d|_t||_t|_dS)Nr%) r-r.ERNIE_SELF_ATTENTION_CLASSESZ_attn_implementationrFroutputset pruned_headsrqrHrJrKr.#s  zErnieAttention.__init__cCst|dkrdSt||jj|jj|j\}}t|jj||j_t|jj||j_t|jj ||j_ t|j j |dd|j _ |jjt||j_|jj|jj|j_ |j ||_dS)Nrr r~)lenrrFrgrirrrlrmrnrrrjunion)rFheadsindexrJrJrK prune_heads+s zErnieAttention.prune_headsFrvc Cs<||||||||}||d|} | f|dd} | S)Nrr )rFr) rFrwrxryrzr{r|r}Z self_outputsattention_outputrrJrJrKrY=s  zErnieAttention.forward)N)NNNNNF) rZr[r\r.rr?r`rr^rrrYrarJrJrHrKr"s$rcs0eZdZfddZejejdddZZS)ErnieIntermediatecsBtt|j|j|_t|jt r6t |j|_ n|j|_ dSr) r-r.rrkr1intermediate_sizer isinstance hidden_actstrr intermediate_act_fnrErHrJrKr.Ws   zErnieIntermediate.__init__rwrPcCs||}||}|Sr)rrrFrwrJrJrKrY_s  zErnieIntermediate.forwardrrJrJrHrKrVs rcs4eZdZfddZejejejdddZZS) ErnieOutputcsBtt|j|j|_tj|j|jd|_t |j |_ dSr) r-r.rrkrr1rr9r:r;r<r=rErHrJrKr.gs zErnieOutput.__init__rcCs&||}||}|||}|SrrrrJrJrKrYms  zErnieOutput.forwardrrJrJrHrKrfs rc steZdZfddZd ejeejeejeejeejeeeejee eejdddZ dd Z Z S) ErnieLayercsrt|j|_d|_t||_|j|_|j|_|jrZ|jsLt|dt|dd|_ t ||_ t ||_ dS)Nr z> should be used as a decoder model if cross attention is addedr&r)r-r.chunk_size_feed_forward seq_len_dimr attentionrpadd_cross_attentionrhcrossattentionr intermediaterrrErHrJrKr.vs   zErnieLayer.__init__NFrvc Cs|dk r|ddnd}|j|||||d} | d} |jrP| dd} | d} n | dd} d} |jr|dk rt|dstd|d|dk r|d dnd}|| ||||||}|d} | |dd} |d} | | } t|j|j|j| }|f| } |jr| | f} | S) Nrf)r}r|rr r(rz'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`r) rrprSrhrrfeed_forward_chunkrr)rFrwrxryrzr{r|r}Zself_attn_past_key_valueZself_attention_outputsrrZpresent_key_valueZcross_attn_present_key_valueZcross_attn_past_key_valueZcross_attention_outputs layer_outputrJrJrKrYsV         zErnieLayer.forwardcCs||}|||}|Sr)rr)rFrZintermediate_outputrrJrJrKrs  zErnieLayer.feed_forward_chunk)NNNNNF) rZr[r\r.r?r`rr^rrrYrrarJrJrHrKrus$  ArcseZdZfddZd ejeejeejeejeejeeeejee ee ee ee e eeje fd ddZ Z S) ErnieEncodercs:t|_tfddtjD|_d|_dS)Ncsg|] }tqSrJ)r).0_rGrJrK sz)ErnieEncoder.__init__..F) r-r.rGrZ ModuleListrangenum_hidden_layerslayergradient_checkpointingrErHrrKr.s  zErnieEncoder.__init__NFT) rwrxryrzr{past_key_valuesrr}output_hidden_states return_dictrPc  Csb| rdnd} |rdnd} |r(|jjr(dnd} |jrJ|jrJ|rJtdd}|rRdnd}t|jD]\}}| rv| |f} |dk r||nd}|dk r||nd}|jr|jr||j |||||||}n||||||||}|d}|r||df7}|r`| |df} |jjr`| |df} q`| r.| |f} | sPt dd ||| | | fDSt ||| | | d S) NrJzZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...Frr(r rfcss|]}|dk r|VqdSrrJ)rvrJrJrK sz'ErnieEncoder.forward..)last_hidden_staterrw attentionscross_attentions) rGrrZtrainingloggerZ warning_once enumeraterZ_gradient_checkpointing_func__call__tupler )rFrwrxryrzr{rrr}rrZall_hidden_statesZall_self_attentionsZall_cross_attentionsZnext_decoder_cacheiZ layer_moduleZlayer_head_maskr|Z layer_outputsrJrJrKrYsx          zErnieEncoder.forward) NNNNNNFFT)rZr[r\r.r?r`rr^rrrr rYrarJrJrHrKrs. rcs0eZdZfddZejejdddZZS) ErniePoolercs*tt|j|j|_t|_dSr)r-r.rrkr1rZTanh activationrErHrJrKr.+s zErniePooler.__init__rcCs(|dddf}||}||}|S)Nr)rr)rFrwZfirst_token_tensor pooled_outputrJrJrKrY0s  zErniePooler.forwardrrJrJrHrKr*s rcs0eZdZfddZejejdddZZS)ErniePredictionHeadTransformcsVtt|j|j|_t|jtr6t |j|_ n|j|_ tj |j|j d|_ dSr) r-r.rrkr1rrrrr transform_act_fnr9r:rErHrJrKr.;s   z%ErniePredictionHeadTransform.__init__rcCs"||}||}||}|Sr)rrr9rrJrJrKrYDs   z$ErniePredictionHeadTransform.forwardrrJrJrHrKr:s rcs,eZdZfddZddZddZZS)ErnieLMPredictionHeadcsLtt||_tj|j|jdd|_t t |j|_ |j |j_ dS)NF)bias) r-r.r transformrrkr1r0decoder Parameterr?rBrrErHrJrKr.Ms   zErnieLMPredictionHead.__init__cCs|j|j_dSr)rrrFrJrJrK _tie_weightsZsz"ErnieLMPredictionHead._tie_weightscCs||}||}|Sr)rrrrJrJrKrY]s  zErnieLMPredictionHead.forward)rZr[r\r.rrYrarJrJrHrKrLs rcs0eZdZfddZejejdddZZS)ErnieOnlyMLMHeadcstt||_dSr)r-r.r predictionsrErHrJrKr.es zErnieOnlyMLMHead.__init__)sequence_outputrPcCs||}|Sr)r)rFrprediction_scoresrJrJrKrYis zErnieOnlyMLMHead.forwardrrJrJrHrKrds rcs$eZdZfddZddZZS)ErnieOnlyNSPHeadcstt|jd|_dSNrf)r-r.rrkr1seq_relationshiprErHrJrKr.ps zErnieOnlyNSPHead.__init__cCs||}|Sr)r)rFrseq_relationship_scorerJrJrKrYts zErnieOnlyNSPHead.forwardrZr[r\r.rYrarJrJrHrKros rcs$eZdZfddZddZZS)ErniePreTrainingHeadscs(tt||_t|jd|_dSr)r-r.rrrrkr1rrErHrJrKr.{s  zErniePreTrainingHeads.__init__cCs||}||}||fSr)rr)rFrrrrrJrJrKrYs  zErniePreTrainingHeads.forwardrrJrJrHrKrzs rc@s$eZdZdZeZdZdZddZdS)ErniePreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. ernieTcCst|tjr:|jjjd|jjd|jdk r|jj nft|tj rz|jjjd|jjd|j dk r|jj|j  n&t|tj r|jj |jj ddS)zInitialize the weightsg)ZmeanZstdNg?)rrrkZweightdataZnormal_rGZinitializer_rangerZzero_r/r#r9Zfill_)rFmodulerJrJrK _init_weightss      z"ErniePreTrainedModel._init_weightsN) rZr[r\r]r! config_classZbase_model_prefixZsupports_gradient_checkpointingrrJrJrJrKrs rc@sleZdZUdZdZeejed<dZ ejed<dZ ejed<dZ ee ejed<dZ ee ejed<dS)ErnieForPreTrainingOutputa] Output type of [`ErnieForPreTraining`]. Args: loss (*optional*, returned when `labels` is provided, `torch.FloatTensor` of shape `(1,)`): Total loss as the sum of the masked language modeling loss and the next sequence prediction (classification) loss. prediction_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). seq_relationship_logits (`torch.FloatTensor` of shape `(batch_size, 2)`): Prediction scores of the next sequence prediction (classification) head (scores of True/False continuation before SoftMax). 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, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs. attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`): Tuple of `torch.FloatTensor` (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. Nlossprediction_logitsseq_relationship_logitsrwr)rZr[r\r]rrr?r^__annotations__rrrwrrrJrJrJrKrs ra? This model inherits from [`PreTrainedModel`]. 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 PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`ErnieConfig`]): 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 (`torch.LongTensor` of shape `({0})`): 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) attention_mask (`torch.FloatTensor` 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 (`torch.LongTensor` 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) task_type_ids (`torch.LongTensor` of shape `({0})`, *optional*): Task type embedding is a special embedding to represent the characteristic of different tasks, such as word-aware pre-training task, structure-aware pre-training task and semantic-aware pre-training task. We assign a `task_type_id` to each task and the `task_type_id` is in the range `[0, config.task_type_vocab_size-1] position_ids (`torch.LongTensor` 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 (`torch.FloatTensor` 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 (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*): Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert `input_ids` indices into associated vectors than the model's internal embedding lookup matrix. output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. z_The bare Ernie Model transformer outputting raw hidden-states without any specific head on top.cseZdZdZdfdd ZddZddZd d Zee d e e e ed deejeejeejeejeejeejeejeejeejeeejeeeeeeeeeeeje fdddZZS) ErnieModela The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added between the self-attention layers, following the architecture described in [Attention is all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin. To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder` argument and `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass. TcsDt|||_t||_t||_|r2t|nd|_| dSr) r-r.rGr"rXrencoderrpooler post_init)rFrGadd_pooling_layerrHrJrKr.s    zErnieModel.__init__cCs|jjSrrXr3rrJrJrKget_input_embeddings*szErnieModel.get_input_embeddingscCs ||j_dSrr)rFrnrJrJrKset_input_embeddings.szErnieModel.set_input_embeddingscCs*|D]\}}|jj|j|qdS)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 N)itemsrrrr)rFZheads_to_prunerrrJrJrK _prune_heads2szErnieModel._prune_headsbatch_size, sequence_length checkpoint output_typerN)rLrxr*rMr'ryrNrzr{rrr}rrrPc Cs`| dk r | n|jj} | dk r | n|jj} |dk r4|n|jj}|jjrZ| dk rP| n|jj} nd} |dk rx|dk rxtdn@|dk r||||}n"|dk r|dd}ntd|\}}|dk r|j n|j }| dk r| ddj dnd}|dkrt j |||f|d}|dkrft |jd rT|jjddd|f}|||}|}nt j|t j|d }|||}|jjr|dk r|\}}}||f}| dkrt j ||d} || }nd}|||jj}|j||||||d }|j|||||| | | | |d }|d}|jdk r&||nd}|sD||f|d dSt|||j|j|j|jdS)a  encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder. encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding. If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`. use_cache (`bool`, *optional*): If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see `past_key_values`). NFzDYou cannot specify both input_ids and inputs_embeds at the same timer(z5You have to specify either input_ids or inputs_embedsrrf)rRr*rQ)rLr'r*rMrNrO) rxryrzr{rrr}rrr )rZ pooler_outputrrwrr)rGr}ruse_return_dictrprrhZ%warn_if_padding_and_no_attention_maskrCrRrr?ZonesrSrXr*rArBrDZget_extended_attention_maskZinvert_attention_maskZ get_head_maskrrrrrrwrr) rFrLrxr*rMr'ryrNrzr{rrr}rrrTZ batch_sizerUrRrOrVrWZextended_attention_maskZencoder_batch_sizeZencoder_sequence_lengthrZencoder_hidden_shapeZencoder_extended_attention_maskZembedding_outputZencoder_outputsrrrJrJrKrY:s+          zErnieModel.forward)T)NNNNNNNNNNNNNN)rZr[r\r]r.rrrrERNIE_INPUTS_DOCSTRINGformatr_CHECKPOINT_FOR_DOCr_CONFIG_FOR_DOCrr?r`rr^rrrrYrarJrJrHrKr sT    rz Ernie Model with two heads on top as done during the pretraining: a `masked language modeling` head and a `next sentence prediction (classification)` head. cseZdZddgZfddZddZddZee d e e e d de eje eje eje eje eje eje eje eje eje ee ee eeeeje fd d dZZS)ErnieForPreTrainingcls.predictions.decoder.biascls.predictions.decoder.weightcs,t|t||_t||_|dSr)r-r.rrrclsrrErHrJrKr.s   zErnieForPreTraining.__init__cCs |jjjSrrrrrrJrJrKget_output_embeddingssz)ErnieForPreTraining.get_output_embeddingscCs||jj_|j|jj_dSrrrrrrFZnew_embeddingsrJrJrKset_output_embeddingss z)ErnieForPreTraining.set_output_embeddingsrrrN) rLrxr*rMr'ryrNlabelsnext_sentence_labelr}rrrPc  Cs| dk r | n|jj} |j|||||||| | | d } | dd\}}|||\}}d}|dk r| dk rt}||d|jj|d}||dd| d}||}| s||f| dd}|dk r|f|S|St|||| j| j dS)a- 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]` next_sentence_label (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair (see `input_ids` docstring) Indices should be in `[0, 1]`: - 0 indicates sequence B is a continuation of sequence A, - 1 indicates sequence B is a random sequence. kwargs (`Dict[str, any]`, *optional*, defaults to `{}`): Used to hide legacy arguments that have been deprecated. Returns: Example: ```python >>> from transformers import AutoTokenizer, ErnieForPreTraining >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("nghuyong/ernie-1.0-base-zh") >>> model = ErnieForPreTraining.from_pretrained("nghuyong/ernie-1.0-base-zh") >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt") >>> outputs = model(**inputs) >>> prediction_logits = outputs.prediction_logits >>> seq_relationship_logits = outputs.seq_relationship_logits ``` N rxr*rMr'ryrNr}rrrfr()rrrrwr) rGrrrr rsr0rrwr)rFrLrxr*rMr'ryrNrrr}rrrrrrr total_lossloss_fctmasked_lm_lossnext_sentence_lossrrJrJrKrYs>1 zErnieForPreTraining.forward) NNNNNNNNNNNN)rZr[r\_tied_weights_keysr.rrrrrrrrrr?r`rrrrYrarJrJrHrKrsB   rzGErnie Model with a `language modeling` head on top for CLM fine-tuning.cseZdZddgZfddZddZddZee d e e e e d deejeejeejeejeejeejeejeejeejeejeeejeeeeeeeeeeeje fd d dZdddZddZZS)ErnieForCausalLMrrcs@t||jstdt|dd|_t||_| dS)NzMIf you want to use `ErnieForCausalLM` as a standalone, add `is_decoder=True.`Fr r-r.rprwarningrrrrrrErHrJrKr.@s    zErnieForCausalLM.__init__cCs |jjjSrrrrJrJrKrMsz&ErnieForCausalLM.get_output_embeddingscCs||jj_|j|jj_dSrrrrJrJrKrQs z&ErnieForCausalLM.set_output_embeddingsrrN)rLrxr*rMr'ryrNrzr{rrrr}rrrPcCs|dk r |n|jj}| dk r d} |j||||||||| | | | ||d}|d}||}d}| dk r|ddddddf}| ddddf} t}||d|jj| d}|s|f|dd}|dk r|f|S|St|||j |j |j |j dS) a encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder. encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the left-to-right language modeling loss (next word prediction). 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 n `[0, ..., config.vocab_size]` past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding. If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`. use_cache (`bool`, *optional*): If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see `past_key_values`). NF) rxr*rMr'ryrNrzr{rrr}rrrr(r rf)rlogitsrrwrr) rGrrrrr rsr0rrrwrr)rFrLrxr*rMr'ryrNrzr{rrrr}rrrrrZlm_lossZshifted_prediction_scoresr rrJrJrKrYUsL0 zErnieForCausalLM.forwardTc Ksv|j}|dkr||}|dk rh|ddjd}|jd|krF|}n|jdd}|dd|df}||||dS)Nrrfr )rLrxrr)rZnew_ones) rFrLrrxr model_kwargsrTZ past_lengthZremove_prefix_lengthrJrJrKprepare_inputs_for_generations z.ErnieForCausalLM.prepare_inputs_for_generationcs.d}|D] }|tfdd|Df7}q|S)NrJc3s"|]}|d|jVqdS)rN)Z index_selectrrR)rZ past_statebeam_idxrJrKrsz2ErnieForCausalLM._reorder_cache..)r)rFrrZreordered_pastZ layer_pastrJrrK_reorder_caches zErnieForCausalLM._reorder_cache)NNNNNNNNNNNNNNN)NNT)rZr[r\r r.rrrrrrrrrrr?r`rrrrrYrrrarJrJrHrKr9s`   Y rz3Ernie Model with a `language modeling` head on top.cseZdZddgZfddZddZddZee d e e e e d d d deejeejeejeejeejeejeejeejeejeejeeeeeeeeeje fdddZdddZZS)ErnieForMaskedLMrrcs@t||jrtdt|dd|_t||_| dS)NzlIf you want to use `ErnieForMaskedLM` make sure `config.is_decoder=False` for bi-directional self-attention.FrrrErHrJrKr.s  zErnieForMaskedLM.__init__cCs |jjjSrrrrJrJrKrsz&ErnieForMaskedLM.get_output_embeddingscCs||jj_|j|jj_dSrrrrJrJrKrs z&ErnieForMaskedLM.set_output_embeddingsrz'paris'g)\(?)rrrZexpected_outputZ expected_lossN)rLrxr*rMr'ryrNrzr{rr}rrrPcCs| dk r | n|jj} |j||||||||| | | | d }|d}||}d}| dk rvt}||d|jj| d}| s|f|dd}|dk r|f|S|St|||j|j dS)a 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]` N) rxr*rMr'ryrNrzr{r}rrrr(rfrrrwr) rGrrrr rsr0rrwr)rFrLrxr*rMr'ryrNrzr{rr}rrrrrr r rrJrJrKrYs< zErnieForMaskedLM.forwardcKs~|j}|d}|jjdkr"tdtj|||jddfgdd}tj|df|jjtj|j d}tj||gdd}||dS)Nrz.The PAD token should be defined for generationr r(r~rQ)rLrx) rrGr2rhr?rZ new_zerosfullrDrR)rFrLrxrrTZeffective_batch_sizeZ dummy_tokenrJrJrKr:s "z.ErnieForMaskedLM.prepare_inputs_for_generation) NNNNNNNNNNNNN)N)rZr[r\r r.rrrrrrrrrrr?r`rrrrYrrarJrJrHrKrsT   >> from transformers import AutoTokenizer, ErnieForNextSentencePrediction >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("nghuyong/ernie-1.0-base-zh") >>> model = ErnieForNextSentencePrediction.from_pretrained("nghuyong/ernie-1.0-base-zh") >>> 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="pt") >>> outputs = model(**encoding, labels=torch.LongTensor([1])) >>> logits = outputs.logits >>> assert logits[0, 0] < logits[0, 1] # next sentence was random ``` rzoThe `next_sentence_label` argument is deprecated and will be removed in a future version, use `labels` instead.Nrr r(rfr) warningswarn FutureWarningpoprGrrrr rsrrwr)rFrLrxr*rMr'ryrNrr}rrkwargsrrZseq_relationship_scoresr r rrJrJrKrYZsD.   z&ErnieForNextSentencePrediction.forward) NNNNNNNNNNN)rZr[r\r.rrrrrrrr?r`rrrrYrarJrJrHrKrKs8  rz Ernie Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks. cseZdZfddZeeddeej eej eej eej eej eej eej eej ee ee ee e e ej e fd ddZZS) ErnieForSequenceClassificationcsdt||j|_||_t||_|jdk r4|jn|j}t ||_ t |j |j|_ |dSr)r-r. num_labelsrGrrclassifier_dropoutr<rr;r=rkr1 classifierrrFrGr$rHrJrKr.s   z'ErnieForSequenceClassification.__init__rNrc  Cs~| dk r | n|jj} |j|||||||| | | d } | d} || } || }d}|dk r:|jjdkr|jdkr|d|j_n4|jdkr|jtj ks|jtj krd|j_nd|j_|jjdkrt }|jdkr|| | }n |||}nN|jjdkrt }||d|j|d}n|jjdkr:t}|||}| sj|f| dd}|dk rf|f|S|St||| j| jd S) a labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). Nrr Z regressionZsingle_label_classificationZmulti_label_classificationr(rfr)rGrrr=r%Z problem_typer#r,r?rDr_r squeezer rsrrrwr)rFrLrxr*rMr'ryrNrr}rrrrrrr rrJrJrKrYsX       "     z&ErnieForSequenceClassification.forward) NNNNNNNNNNN)rZr[r\r.rrrrr?r`rrrrrYrarJrJrHrKr"s6  r"z Ernie Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks. cseZdZfddZeedeee e dd e e j e e j e e j e e j e e j e e j e e j e e j e ee ee eeee j e fd ddZZS) ErnieForMultipleChoicecsTt|t||_|jdk r&|jn|j}t||_t |j d|_ | dS)Nr ) r-r.rrr$r<rr;r=rkr1r%rr&rHrJrKr. s   zErnieForMultipleChoice.__init__z(batch_size, num_choices, sequence_lengthrNrc  Csv| dk r | n|jj} |dk r&|jdn|jd} |dk rJ|d|dnd}|dk rh|d|dnd}|dk r|d|dnd}|dk r|d|dnd}|dk r|d|d|dnd}|j|||||||| | | d } | d}||}||}|d| }d}|dk r2t}|||}| sb|f| dd}|dk r^|f|S|St ||| j | j dS)aJ 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) Nr r(rrrfr) rGrrrsrCrr=r%r rrwr)rFrLrxr*rMr'ryrNrr}rrZ num_choicesrrrZreshaped_logitsrr rrJrJrKrY-sN      zErnieForMultipleChoice.forward) NNNNNNNNNNN)rZr[r\r.rrrrrrrrr?r`rrrrYrarJrJrHrKr(s@ r(z Ernie 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. cseZdZfddZeeddeej eej eej eej eej eej eej eej ee ee ee e e ej e fd ddZZS) ErnieForTokenClassificationcsbt||j|_t|dd|_|jdk r2|jn|j}t||_ t |j |j|_ | dSNFr)r-r.r#rrr$r<rr;r=rkr1r%rr&rHrJrKr.s  z$ErnieForTokenClassification.__init__rNrc  Cs| dk r | n|jj} |j|||||||| | | d } | d} || } || }d}|dk rzt}||d|j|d}| s|f| dd}|dk r|f|S|St||| j | j dS)z labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`. Nrrr(rfr) rGrrr=r%r rsr#rrwr)rFrLrxr*rMr'ryrNrr}rrrrrrr rrJrJrKrYs:   z#ErnieForTokenClassification.forward) NNNNNNNNNNN)rZr[r\r.rrrrr?r`rrrrrYrarJrJrHrKr)xs6  r)z Ernie 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`). cseZdZfddZeeddeej eej eej eej eej eej eej eej eej ee ee ee e e ej e fd ddZZS) ErnieForQuestionAnsweringcs@t||j|_t|dd|_t|j|j|_| dSr*) r-r.r#rrrrkr1 qa_outputsrrErHrJrKr.s  z"ErnieForQuestionAnswering.__init__rN) rLrxr*rMr'ryrNstart_positions end_positionsr}rrrPc  CsR| dk r | n|jj} |j|||||||| | | d } | d}||}|jddd\}}|d}|d}d}|dk r | dk r t|dkr|d}t| dkr| d} |d}| d|}| d|} t |d}|||}||| }||d}| s<||f| dd}|dk r8|f|S|St |||| j | j d S) a start_positions (`torch.LongTensor` 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 (`torch.LongTensor` 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. Nrrr r(r~)Z ignore_indexrf)r start_logits end_logitsrwr)rGrrr,splitr'rrrCclampr rrwr)rFrLrxr*rMr'ryrNr-r.r}rrrrrr/r0r Z ignored_indexr Z start_lossZend_lossrrJrJrKrYsR           z!ErnieForQuestionAnswering.forward) NNNNNNNNNNNN)rZr[r\r.rrrrr?r`rrrrrYrarJrJrHrKr+s: r+)Nr]rrZ dataclassesrtypingrrrrr?Ztorch.utils.checkpointrZtorch.nnrr r Z activationsr Zmodeling_outputsr rrrrrrrrZmodeling_utilsrZ pytorch_utilsrrrutilsrrrrrrZconfiguration_ernier!Z get_loggerrZrrrModuler"rbrrrrrrrrrrrrrrrZERNIE_START_DOCSTRINGrrrrrrr"r(r)r+rJrJrJrKs   ,    M4W^   !77mqe\ZH