U 4Af @sdZddlZddlZddlZddlmZddlmZmZm Z m Z ddl Z ddl Z ddl m Z ddlmZmZmZddlmZdd lmZmZmZmZmZmZmZmZdd 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/dZ0ddZ1Gddde j2Z3Gddde j2Z4Gddde j2Z5Gddde j2Z6Gddde j2Z7Gddde j2Z8Gdd d e j2Z9Gd!d"d"e j2Z:Gd#d$d$e j2Z;Gd%d&d&e j2ZGd+d,d,e j2Z?Gd-d.d.e j2Z@Gd/d0d0e j2ZAGd1d2d2eZBeGd3d4d4e$ZCd5ZDd6ZEe&d7eDGd8d9d9eBZFe&d:eDGd;d<dd?d?eBZHe&d@eDGdAdBdBeBZIe&dCeDGdDdEdEeBZJe&dFeDGdGdHdHeBZKe&dIeDGdJdKdKeBZLe&dLeDGdMdNdNeBZMdS)OzPyTorch Nezha model.N) dataclass)ListOptionalTupleUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN))BaseModelOutputWithPastAndCrossAttentions,BaseModelOutputWithPoolingAndCrossAttentionsMaskedLMOutputMultipleChoiceModelOutputNextSentencePredictorOutputQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)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) NezhaConfigzsijunhe/nezha-cn-baser c Cszddl}ddl}ddl}Wn tk r<tdYnXtj|}t d||j |}g}g} |D]@\} } t d| d| |j || } | | | | qrt|| D]\} } | d} tdd | Drt d d| q|} | D]}|d |r&|d |}n|g}|dd ksH|ddkrTt| d} n|ddksp|ddkr|t| d} nz|ddkrt| d} n`|ddkrt| d} nFzt| |d} Wn2tk rt d d| YqYnXt|dkrt|d}| |} q|dddkr:t| d} n|d krN|| } z,| j| jkrxtd| jd| jdWn<tk r}z|j| j| jf7_W5d}~XYnXt d| t| | _q|S)z'Load tf checkpoints in a pytorch model.rNzLoading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see https://www.tensorflow.org/install/ for installation instructions.z&Converting TensorFlow checkpoint from zLoading TF weight z with shape /css|]}|dkVqdS))Zadam_vZadam_mZAdamWeightDecayOptimizerZAdamWeightDecayOptimizer_1Z global_stepN).0nr"r"W/tmp/pip-unpacked-wheel-zw5xktn0/transformers/models/deprecated/nezha/modeling_nezha.py Wsz+load_tf_weights_in_nezha..z Skipping z [A-Za-z]+_\d+z_(\d+)kernelgammaweightZ output_biasbetabiasZoutput_weightsZsquad classifierriZ _embeddingszPointer shape z and array shape z mismatchedzInitialize PyTorch weight )reZnumpyZ tensorflow ImportErrorloggererrorospathabspathinfoZtrainZlist_variablesZ load_variableappendzipsplitanyjoin fullmatchgetattrAttributeErrorlenint transposeshape ValueErrorAssertionErrorargstorchZ from_numpydata)modelconfigZtf_checkpoint_pathr.nptfZtf_pathZ init_varsnamesZarraysnamerAarraypointerZm_nameZ scope_namesnumer"r"r%load_tf_weights_in_nezha:sx                  rQcs*eZdZdZdfdd ZddZZS)NezhaRelativePositionsEncodingz3Implement the Functional Relative Position EncodingcsHt|dd}t|}||||}|t|}t|| |}||} t||} tjd|tj d d} t td|d t d |} t| | | dddddf<t| | | dddddf<| d} tjjj| |d }t|| }t| }||||}|jd|d d dS) Nr-rrdtypeg@)Z num_classespositions_encodingF persistent)super__init__rEZarangerepeatviewtclampzerosZint64floatZ unsqueezeexpmathlogsincosr functionalZone_hotmatmullistsizer6register_buffer)selflengthdepthmax_relative_position vocab_sizeZ range_vecZ range_matZ distance_matZdistance_mat_clippedZ final_matZembeddings_tablepositionZdiv_termZflat_relative_positions_matrixZ!one_hot_relative_positions_matrixrWZmy_shape __class__r"r%r[s,    (        z'NezhaRelativePositionsEncoding.__init__cCs|jd|d|ddfSN)rW)rlrmr"r"r%forwardsz&NezhaRelativePositionsEncoding.forward)rS)__name__ __module__ __qualname____doc__r[ru __classcell__r"r"rrr%rRsrRcsJeZdZdZfddZdeejeejeejej dddZ Z S) NezhaEmbeddingsz=Construct the embeddings from word and token_type embeddings.cs|ttj|j|j|jd|_t|j|j|_ tj |j|j d|_ t |j |_|jdtjd|jftjddddS)N) padding_idxZepstoken_type_idsrrTFrX)rZr[r Embeddingrp hidden_size pad_token_idword_embeddingsZtype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutrkrEr`max_position_embeddingslongrlrHrrr"r%r[s zNezhaEmbeddings.__init__N) input_idsr~ inputs_embedsreturnc Cs|dk r|}n|dd}|d}|dkr<||}|dkrt|drz|jddd|f}||d|}|}ntj|tj|jd}| |}||} | | } | | } | S)NrVrr~rrUdevice) rjrhasattrr~expandrEr`rrrrr) rlrr~r input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expandedr embeddingsr"r"r%rus"      zNezhaEmbeddings.forward)NNN) rvrwrxryr[rrEZ LongTensor FloatTensorTensorrurzr"r"rrr%r{s r{c seZdZfddZejejdddZd ejeejeejeejeejee e ejee e ejdd d Z Z S) NezhaSelfAttentioncst|j|jdkr4td|jd|jd|j|_t|j|j|_|j|j|_t |j|j|_ t |j|j|_ t |j|j|_ t |j|_t|j|j|jd|_|j|_dS)NrzThe hidden size (z6) is not a multiple of the number of attention heads ())rmrnro)rZr[rnum_attention_headsrBr?attention_head_size all_head_sizerLinearquerykeyvaluerZattention_probs_dropout_probrrRrrorelative_positions_encoding is_decoderrrrr"r%r[s$ zNezhaSelfAttention.__init__)xrcCs6|dd|j|jf}||}|ddddS)NrVrr-r)rjrrr]permute)rlrZ new_x_shaper"r"r%transpose_for_scoress z'NezhaSelfAttention.transpose_for_scoresNF hidden_statesattention_mask head_maskencoder_hidden_statesencoder_attention_maskpast_key_valueoutput_attentionsrc"Cs||}|dk } | r4|dk r4|d} |d} |}n| r^|||} |||} |}nv|dk r|||} |||} tj|d| gdd} tj|d| gdd} n |||} |||} ||} |jr| | f}t| | dd} | \}}}}| |}| dddd}| ||||j}t|| ddd}| ||||}| dddd}| |} | t|j} |dk r| |} tjj| dd}||}|dk r||}t|| }| |}| dddd}| ||||}t||}| ||||j}| dddd}||}| dddd }| dd|jf} | | }|r||fn|f}!|jr|!|f}!|!S)Nrrr-ZdimrVr)rrrrrEcatrrhr@rjrr contiguousr]rrcsqrtrrgZsoftmaxrr)"rlrrrrrrrZmixed_query_layerZis_cross_attentionZ key_layerZ value_layerZ query_layerZattention_scores batch_sizerZfrom_seq_lengthZ to_seq_lengthZrelations_keysZ query_layer_tZ query_layer_rZkey_position_scoresZkey_position_scores_rZkey_position_scores_r_tZattention_probsZ context_layerZrelations_valuesZattention_probs_tZattentions_probs_rZvalue_position_scoresZvalue_position_scores_rZvalue_position_scores_r_tZnew_context_layer_shapeoutputsr"r"r%rus            zNezhaSelfAttention.forward)NNNNNF) rvrwrxr[rErrrrrboolrurzr"r"rrr%rs$ rcs4eZdZfddZejejejdddZZS)NezhaSelfOutputcsBtt|j|j|_tj|j|jd|_t|j |_ dSNr}) rZr[rrrdenserrrrrrrrr"r%r[^s zNezhaSelfOutput.__init__r input_tensorrcCs&||}||}|||}|Srtrrrrlrrr"r"r%ruds  zNezhaSelfOutput.forwardrvrwrxr[rErrurzr"r"rrr%r]s rc steZdZfddZddZd ejeejeejeejeejee e ejee e ejddd Z Z S) NezhaAttentioncs*tt||_t||_t|_dSrt)rZr[rrlroutputset pruned_headsrrrr"r%r[ls   zNezhaAttention.__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)Nrrr)r>rrlrrrrrrrrrrunion)rlheadsindexr"r"r% prune_headsrs zNezhaAttention.prune_headsNFrc Cs<||||||||}||d|} | f|dd} | S)Nrr)rlr) rlrrrrrrrZ self_outputsattention_outputrr"r"r%rus  zNezhaAttention.forward)NNNNNF) rvrwrxr[rrErrrrrrurzr"r"rrr%rks$ rcs0eZdZfddZejejdddZZS)NezhaIntermediatecsBtt|j|j|_t|jt r6t |j|_ n|j|_ dSrt) rZr[rrrintermediate_sizer isinstance hidden_actstrr intermediate_act_fnrrrr"r%r[s   zNezhaIntermediate.__init__rrcCs||}||}|Srt)rrrlrr"r"r%rus  zNezhaIntermediate.forwardrr"r"rrr%rs rcs4eZdZfddZejejejdddZZS) NezhaOutputcsBtt|j|j|_tj|j|jd|_t |j |_ dSr) rZr[rrrrrrrrrrrrrr"r%r[s zNezhaOutput.__init__rcCs&||}||}|||}|Srtrrr"r"r%rus  zNezhaOutput.forwardrr"r"rrr%rs rc steZdZfddZd ejeejeejeejeejeeeejee eejdddZ dd Z Z S) NezhaLayercsnt|j|_d|_t||_|j|_|j|_|jrV|jsLt|dt||_ t ||_ t ||_ dS)Nrz> should be used as a decoder model if cross attention is added)rZr[chunk_size_feed_forward seq_len_dimr attentionradd_cross_attentionrBcrossattentionr intermediaterrrrrr"r%r[s    zNezhaLayer.__init__NFrc 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) Nr-)rrrrrVrz'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`r) rrrrBrrfeed_forward_chunkrr)rlrrrrrrrZself_attn_past_key_valueZself_attention_outputsrrZpresent_key_valueZcross_attn_present_key_valueZcross_attn_past_key_valueZcross_attention_outputs layer_outputr"r"r%rusV         zNezhaLayer.forwardcCs||}|||}|Srt)rr)rlrZintermediate_outputrr"r"r%r s  zNezhaLayer.feed_forward_chunk)NNNNNF) rvrwrxr[rErrrrrrurrzr"r"rrr%rs$  ArcseZdZfddZd ejeejeejeejeejeeeejee ee ee ee e eeje fd ddZ Z S) NezhaEncodercs:t|_tfddtjD|_d|_dS)Ncsg|] }tqSr")r)r#_rHr"r% sz)NezhaEncoder.__init__..F) rZr[rHrZ ModuleListrangenum_hidden_layerslayergradient_checkpointingrrrrr%r[s  zNezhaEncoder.__init__NFT) rrrrrpast_key_values use_cacheroutput_hidden_states return_dictrc  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) Nr"zZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...FrrVrr-css|]}|dk r|VqdSrtr")r#vr"r"r%r&Xsz'NezhaEncoder.forward..)last_hidden_staterr attentionscross_attentions) rHrrZtrainingr0Z warning_once enumeraterZ_gradient_checkpointing_func__call__tupler )rlrrrrrrrrrrZall_hidden_statesZall_self_attentionsZall_cross_attentionsZnext_decoder_cacheiZ layer_moduleZlayer_head_maskrZ layer_outputsr"r"r%rusx          zNezhaEncoder.forward) NNNNNNFFT)rvrwrxr[rErrrrrrr rurzr"r"rrr%rs. rcs0eZdZfddZejejdddZZS) NezhaPoolercs*tt|j|j|_t|_dSrt)rZr[rrrrZTanh activationrrrr"r%r[ms zNezhaPooler.__init__rcCs(|dddf}||}||}|S)Nr)rr)rlrZfirst_token_tensor pooled_outputr"r"r%rurs  zNezhaPooler.forwardrr"r"rrr%rls rcs0eZdZfddZejejdddZZS)NezhaPredictionHeadTransformcsVtt|j|j|_t|jtr6t |j|_ n|j|_ tj |j|j d|_ dSr) rZr[rrrrrrrr transform_act_fnrrrrrr"r%r[|s   z%NezhaPredictionHeadTransform.__init__rcCs"||}||}||}|Srt)rrrrr"r"r%rus   z$NezhaPredictionHeadTransform.forwardrr"r"rrr%r{s rcs,eZdZfddZddZddZZS)NezhaLMPredictionHeadcsLtt||_tj|j|jdd|_t t |j|_ |j |j_ dS)NF)r+) rZr[r transformrrrrpdecoder ParameterrEr`r+rrrr"r%r[s   zNezhaLMPredictionHead.__init__cCs|j|j_dSrt)r+rrlr"r"r% _tie_weightssz"NezhaLMPredictionHead._tie_weightscCs||}||}|Srt)rrrr"r"r%rus  zNezhaLMPredictionHead.forward)rvrwrxr[rrurzr"r"rrr%rs rcs0eZdZfddZejejdddZZS)NezhaOnlyMLMHeadcstt||_dSrt)rZr[r predictionsrrrr"r%r[s zNezhaOnlyMLMHead.__init__)sequence_outputrcCs||}|Srt)r)rlrprediction_scoresr"r"r%rus zNezhaOnlyMLMHead.forwardrr"r"rrr%rs rcs$eZdZfddZddZZS)NezhaOnlyNSPHeadcstt|jd|_dSNr-)rZr[rrrseq_relationshiprrrr"r%r[s zNezhaOnlyNSPHead.__init__cCs||}|Srt)r)rlrseq_relationship_scorer"r"r%rus zNezhaOnlyNSPHead.forwardrvrwrxr[rurzr"r"rrr%rs rcs$eZdZfddZddZZS)NezhaPreTrainingHeadscs(tt||_t|jd|_dSr)rZr[rrrrrrrrrr"r%r[s  zNezhaPreTrainingHeads.__init__cCs||}||}||fSrt)rr)rlrrrrr"r"r%rus  zNezhaPreTrainingHeads.forwardrr"r"rrr%rs rc@s(eZdZdZeZeZdZdZ ddZ dS)NezhaPreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. nezhaTcCst|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?)rrrr)rFZnormal_rHZinitializer_ranger+Zzero_rr|rZfill_)rlmoduler"r"r% _init_weightss      z"NezhaPreTrainedModel._init_weightsN) rvrwrxryr config_classrQZload_tf_weightsZbase_model_prefixZsupports_gradient_checkpointingrr"r"r"r%rs rc@sleZdZUdZdZeejed<dZ ejed<dZ ejed<dZ ee ejed<dZ ee ejed<dS)NezhaForPreTrainingOutputa] Output type of [`NezhaForPreTraining`]. 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_logitsrr)rvrwrxryr rrEr__annotations__r r rrrr"r"r"r%r s r a? 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 ([`NezhaConfig`]): 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) 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 Nezha Model transformer outputting raw hidden-states without any specific head on top.cseZdZdZdfdd ZddZddZd d Zee d e e e ed deejeejeejeejeejeejeejeeejeeeeeeeeeeeje fd ddZZS) NezhaModela 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|_| dSrt) rZr[rHr{rrencoderrpooler post_init)rlrHadd_pooling_layerrrr"r%r[Os    zNezhaModel.__init__cCs|jjSrtrrrr"r"r%get_input_embeddings[szNezhaModel.get_input_embeddingscCs ||j_dSrtr)rlrr"r"r%set_input_embeddings^szNezhaModel.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)rlZheads_to_prunerrr"r"r% _prune_headsaszNezhaModel._prune_headsbatch_size, sequence_length checkpoint output_typer N) rrr~rrrrrrrrrrc  CsZ| 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}| s>||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 timerVz5You have to specify either input_ids or inputs_embedsrr-)rr~r)rr~r) rrrrrrrrrr)rZ pooler_outputrrrr)rHrruse_return_dictrrrBZ%warn_if_padding_and_no_attention_maskrjrrArEZonesrrr~rr`rZget_extended_attention_maskZinvert_attention_maskZ get_head_maskrrrrrrrr)rlrrr~rrrrrrrrrrrrrZpast_key_values_lengthrrZextended_attention_maskZencoder_batch_sizeZencoder_sequence_lengthrZencoder_hidden_shapeZencoder_extended_attention_maskZembedding_outputZencoder_outputsrrr"r"r%ruis)          zNezhaModel.forward)T) NNNNNNNNNNNN)rvrwrxryr[rrrrNEZHA_INPUTS_DOCSTRINGformatr_CHECKPOINT_FOR_DOCr_CONFIG_FOR_DOCrrErrrrrrrurzr"r"rrr%r>sL    rz Nezha Model with two heads on top as done during the pretraining: a `masked language modeling` head and a `next sentence prediction (classification)` head. cseZdZdgZfddZddZddZee de e e d de eje eje eje eje eje eje eje ee ee eeeeje fd d d ZZS)NezhaForPreTrainingcls.predictions.decodercs,t|t||_t||_|dSrt)rZr[rrrclsrrrrr"r%r[s   zNezhaForPreTraining.__init__cCs |jjjSrtr$rrrr"r"r%get_output_embeddingssz)NezhaForPreTraining.get_output_embeddingscCs||jj_|j|jj_dSrtr$rrr+rlZnew_embeddingsr"r"r%set_output_embeddingss z)NezhaForPreTraining.set_output_embeddingsrrr N) rrr~rrlabelsnext_sentence_labelrrrrc  Cs| dk r | n|jj} |j||||||| | d} | dd\} } || | \}}d}|dk r|dk rt}||d|jj|d}||dd|d}||}| s||f| dd}|dk r|f|S|St|||| j| j dS)a! labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]` next_sentence_label (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair (see `input_ids` docstring) Indices should be in `[0, 1]`: - 0 indicates sequence B is a continuation of sequence A, - 1 indicates sequence B is a random sequence. kwargs (`Dict[str, any]`, optional, defaults to *{}*): Used to hide legacy arguments that have been deprecated. Returns: Example: ```python >>> from transformers import AutoTokenizer, NezhaForPreTraining >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("sijunhe/nezha-cn-base") >>> model = NezhaForPreTraining.from_pretrained("sijunhe/nezha-cn-base") >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt") >>> outputs = model(**inputs) >>> prediction_logits = outputs.prediction_logits >>> seq_relationship_logits = outputs.seq_relationship_logits ``` Nrr~rrrrrr-rV)r r r rr) rHrrr$r r]rpr rr)rlrrr~rrr+r,rrrrrrrr total_lossloss_fctmasked_lm_lossnext_sentence_lossrr"r"r%ru s:/ zNezhaForPreTraining.forward) NNNNNNNNNN)rvrwrx_tied_weights_keysr[r&r)rrrrr r!rrErrrrrurzr"r"rrr%r"s:   r"z3Nezha Model with a `language modeling` head on top.cseZdZdgZfddZddZddZee de e e e d deejeejeejeejeejeejeejeejeeeeeeeeeje fd d d ZdddZZS)NezhaForMaskedLMr#cs@t||jrtdt|dd|_t||_| dS)NzlIf you want to use `NezhaForMaskedLM` make sure `config.is_decoder=False` for bi-directional self-attention.Fr) rZr[rr0warningrrrr$rrrrr"r%r[`s  zNezhaForMaskedLM.__init__cCs |jjjSrtr%rr"r"r%r&osz&NezhaForMaskedLM.get_output_embeddingscCs||jj_|j|jj_dSrtr'r(r"r"r%r)rs z&NezhaForMaskedLM.set_output_embeddingsrrN) rrr~rrrrr+rrrrc  Cs| dk r | n|jj} |j|||||||| | | d } | d} || }d}|dk rrt}||d|jj|d}| s|f| dd}|dk r|f|S|St||| j| j dS)a labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]` N) rr~rrrrrrrrrVr-r logitsrr) rHrrr$r r]rprrr)rlrrr~rrrrr+rrrrrrr0r/rr"r"r%ruvs8  zNezhaForMaskedLM.forwardcKs~|j}|d}|jjdkr"tdtj|||jddfgdd}tj|df|jjtj|j d}tj||gdd}||dS)Nrz.The PAD token should be defined for generationrrVrr)rr) rArHrrBrErZ new_zerosfullrr)rlrrZ model_kwargsrZeffective_batch_sizeZ dummy_tokenr"r"r%prepare_inputs_for_generations "z.NezhaForMaskedLM.prepare_inputs_for_generation) NNNNNNNNNNN)N)rvrwrxr2r[r&r)rrrrr rr!rrErrrrrur9rzr"r"rrr%r3\sH  7r3zKNezha Model with a `next sentence prediction (classification)` head on top.cseZdZfddZeedeee dd e e j e e j e e j e e j e e j e e j e e e e e e eee j efd ddZZS) NezhaForNextSentencePredictioncs,t|t||_t||_|dSrt)rZr[rrrr$rrrrr"r%r[s   z'NezhaForNextSentencePrediction.__init__rr*N rrr~rrr+rrrrc  Ksd| krtdt| d}| dk r*| n|jj} |j|||||||| d} | d} || } d}|dk rt}|| dd| d}| s| f| dd}|dk r|f|S|St || | j | j dS) a labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair (see `input_ids` docstring). Indices should be in `[0, 1]`: - 0 indicates sequence B is a continuation of sequence A, - 1 indicates sequence B is a random sequence. Returns: Example: ```python >>> from transformers import AutoTokenizer, NezhaForNextSentencePrediction >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("sijunhe/nezha-cn-base") >>> model = NezhaForNextSentencePrediction.from_pretrained("sijunhe/nezha-cn-base") >>> prompt = "In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced." >>> next_sentence = "The sky is blue due to the shorter wavelength of blue light." >>> encoding = tokenizer(prompt, next_sentence, return_tensors="pt") >>> outputs = model(**encoding, labels=torch.LongTensor([1])) >>> logits = outputs.logits >>> assert logits[0, 0] < logits[0, 1] # next sentence was random ``` r,zoThe `next_sentence_label` argument is deprecated and will be removed in a future version, use `labels` instead.Nr-rrVr-r6) warningswarn FutureWarningpoprHrrr$r r]rrr)rlrrr~rrr+rrrkwargsrrZseq_relationship_scoresr1r/rr"r"r%rus@,   z&NezhaForNextSentencePrediction.forward) NNNNNNNNN)rvrwrxr[rrrrrr!rrErrrrrurzr"r"rrr%r:s0  r:z Nezha Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks. cseZdZfddZeedeee e dd 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) NezhaForSequenceClassificationcsdt||j|_||_t||_|jdk r4|jn|j}t ||_ t |j |j|_ |dSrt)rZr[ num_labelsrHrrclassifier_dropoutrrrrrrr,rrlrHrCrrr"r%r[0s   z'NezhaForSequenceClassification.__init__rrNr;c  Csz| dk r | n|jj} |j|||||||| d} | d} || } || } d} |dk r6|jjdkr|jdkrxd|j_n4|jdkr|jtj ks|jtj krd|j_nd|j_|jjdkrt }|jdkr|| | } n || |} nN|jjdkrt }|| d|j|d} n|jjdkr6t}|| |} | sf| f| dd}| dk rb| 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). Nr-rZ regressionZsingle_label_classificationZmulti_label_classificationrVr-r6)rHrrrr,Z problem_typerBrUrErr?r squeezer r]rrrr)rlrrr~rrr+rrrrrr7r r/rr"r"r%ru?sT       "     z&NezhaForSequenceClassification.forward) NNNNNNNNN)rvrwrxr[rrrrr rr!rrErrrrrurzr"r"rrr%rA(s8  rAz Nezha 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 ee ee eeee j e fd ddZZS) NezhaForMultipleChoicecsTt|t||_|jdk r&|jn|j}t||_t |j d|_ | dS)Nr) rZr[rrrCrrrrrrr,rrDrrr"r%r[s   zNezhaForMultipleChoice.__init__z(batch_size, num_choices, sequence_lengthrNr;c  Csp| dk r | n|jj} |dk r&|jdn|jd} |dk rJ|d|dnd}|dk rh|d|dnd}|dk r|d|dnd}|dk r|d|d|dnd}|j|||||||| d} | d} t| j|| } || } t| jt| | d| }d}|dk r,t }|||}| s\|f| dd}|dk rX|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) NrrVrr-r-r6) rHrrAr]rjrprintrr,r rrr)rlrrr~rrr+rrrZ num_choicesrrr7Zreshaped_logitsr r/rr"r"r%rusN        zNezhaForMultipleChoice.forward) NNNNNNNNN)rvrwrxr[rrrrr rr!rrErrrrrurzr"r"rrr%rFs8 rFz Nezha 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. cseZdZfddZeedeee e dd 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) NezhaForTokenClassificationcsbt||j|_t|dd|_|jdk r2|jn|j}t||_ t |j |j|_ | dSNFr4)rZr[rBrrrCrrrrrrr,rrDrrr"r%r[s  z$NezhaForTokenClassification.__init__rrNr;c  Cs| dk r | n|jj} |j|||||||| d} | d} || } || } d} |dk rvt}|| 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]`. Nr-rrVr-r6) rHrrrr,r r]rBrrr)rlrrr~rrr+rrrrrr7r r/rr"r"r%rus6   z#NezhaForTokenClassification.forward) NNNNNNNNN)rvrwrxr[rrrrr rr!rrErrrrrurzr"r"rrr%rHs8  rHz Nezha 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`). 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 ee ee eeee j e fd ddZZS) NezhaForQuestionAnsweringcs@t||j|_t|dd|_t|j|j|_| dSrI) rZr[rBrrrrr qa_outputsrrrrr"r%r[>s  z"NezhaForQuestionAnswering.__init__rrN) rrr~rrstart_positions end_positionsrrrrc  CsN| dk r | n|jj} |j||||||| | d} | d} || } | jddd\}}|d}|d}d}|dk r|dk rt|dkr|d}t|dkr|d}|d}| d|}| d|}t |d}|||}|||}||d}| s8||f| dd}|dk r4|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. Nr-rrrVr)Z ignore_indexr-)r  start_logits end_logitsrr)rHrrrKr8rErr>rjr_r rrr)rlrrr~rrrLrMrrrrrr7rNrOr.Z ignored_indexr/Z start_lossZend_lossrr"r"r%ruHsN           z!NezhaForQuestionAnswering.forward) NNNNNNNNNN)rvrwrxr[rrrrr rr!rrErrrrrurzr"r"rrr%rJ6s< rJ)Nryrcr2r<Z dataclassesrtypingrrrrrEZtorch.utils.checkpointrZtorch.nnrr r Z activationsr Zmodeling_outputsr rrrrrrrZmodeling_utilsrZ pytorch_utilsrrrutilsrrrrrrZconfiguration_nezhar Z get_loggerrvr0r r!rQModulerRr{rrrrrrrrrrrrrrr ZNEZHA_START_DOCSTRINGrrr"r3r:rArFrHrJr"r"r"r%s   (    I 31V]   !-.fg`\UH