o ߥi@sdZddlZddlmZddlmZmZddlZddlm Z ddl Zddl m Z ddl mZddlmZmZmZmZddlmZdd lmZmZdd lmZdd lmZdd lmZdd lm Z ddl!m"Z"ddl#m$Z$e Z%Gddde j&Z'Gddde j&Z(Gddde j&Z)Gddde j&Z*Gddde j&Z+Gddde j&Z,Gddde j&Z-Gdd d e j&Z.Gd!d"d"e j&Z/Gd#d$d$eeZ0eGd%d&d&eZ1ej2ej3ej4d'Gd(d)d)e0Z5dS)*zRPyTorch StructBERT model. mainly copied from :module:`~transformers.modeling_bert`N) dataclass)OptionalUnion)version)ACT2FN)PreTrainedModelapply_chunking_to_forward find_pruneable_heads_and_indicesprune_linear_layer)Models)Model TorchModel)MODELS)AttentionBackboneModelOutput)Tasks) get_logger)parse_labels_in_order) SbertConfigcs6eZdZdZfddZ      d ddZZS) SbertEmbeddingszGConstruct the embeddings from word, position and token_type embeddings.csttj|j|j|jd|_t|j|j|_ t|j |j|_ tj |j|j d|_ t|j|_t|dd|_|dt|jdttjtdkri|jdtj|jtj|jjd d d dSdS) N) padding_idxepsposition_embedding_typeabsolute position_ids)rz1.6.0token_type_idsdtypedeviceF) persistent)super__init__nn Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingstype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutgetattrrregister_buffertorcharangeexpandrparse __version__zerosrsizelongr selfconfig __class__]/home/ubuntu/.local/lib/python3.10/site-packages/modelscope/models/nlp/structbert/backbone.pyr#0sF  zSbertEmbeddings.__init__NrFcCs|dur |}n|dd}|d}|dur&|jdd|||f}|durPt|drE|jddd|f} | |d|} | }n tj|tj|jjd}|durY| |}| |} || } |j dkrp| |} | | 7} | | } || } |s~| S| |fS)Nrrrrrr)r;rhasattrrr7r5r:r<r r)r-rr+r.r2)r> input_idsrr inputs_embedspast_key_values_lengthreturn_inputs_embeds input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expandedr- embeddingsr+rBrBrCforwardQsD         zSbertEmbeddings.forward)NNNNrF)__name__ __module__ __qualname____doc__r#rN __classcell__rBrBr@rCr-s "rc:eZdZfddZddZ      d ddZZS) SbertSelfAttentioncst|j|jdkrt|dstd|jd|jd|j|_t|j|j|_|j|j|_t |j|j|_ t |j|j|_ t |j|j|_ t |j|_t|dd|_|jdksf|jd krw|j|_t d |jd |j|_|j|_dS) Nrembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ()rr relative_keyrelative_key_queryr)r"r#r'num_attention_headsrD ValueErrorintattention_head_size all_head_sizer$Linearquerykeyvaluer0attention_probs_dropout_probr2r3rr*r%distance_embedding is_decoderr=r@rBrCr#s:    zSbertSelfAttention.__init__cCs6|dd|j|jf}|j|}|ddddS)NrrrZr)r;r[r^viewpermute)r>x new_x_shaperBrBrCtranspose_for_scoress  z'SbertSelfAttention.transpose_for_scoresNFcCs||}|du} | r|dur|d} |d} |}nP| r/|||} |||} |}n;|durZ|||} |||} tj|d| gdd} tj|d| gdd} n|||} |||} ||} |jrv| | f}t| | dd} |j dks|j dkr| d}tj |tj |j d dd}tj |tj |j d dd}||}|||jd}|j| jd }|j dkrtd | |}| |} n|j dkrtd | |}td | |}| ||} | t|j} |dur| |} tjdd| }||}|dur||}t|| }|dddd }| dd|jf}|j|}|r9||fn|f}|jrE||f}|S)NrrrZdimrrXrYr)rzbhld,lrd->bhlrzbhrd,lrd->bhlrrg)rarlrbrcr5catrfmatmul transposerr;r6r<r rhrer*toreinsummathsqrtr^r$Softmaxr2ri contiguousr_)r> hidden_statesattention_mask head_maskencoder_hidden_statesencoder_attention_maskpast_key_valueoutput_attentionsmixed_query_layeris_cross_attention key_layer value_layer query_layerattention_scoresrJposition_ids_lposition_ids_rdistancepositional_embeddingrelative_position_scoresrelative_position_scores_queryrelative_position_scores_keyattention_probs context_layernew_context_layer_shapeoutputsrBrBrCrNs               zSbertSelfAttention.forwardNNNNNF)rOrPrQr#rlrNrSrBrBr@rCrUs  rUc$eZdZfddZddZZS)SbertSelfOutputcsBtt|j|j|_tj|j|jd|_t|j |_ dSNr) r"r#r$r`r'denser.r/r0r1r2r=r@rBrCr# zSbertSelfOutput.__init__cC&||}||}|||}|SNrr2r.r>ry input_tensorrBrBrCrN   zSbertSelfOutput.forwardrOrPrQr#rNrSrBrBr@rCr rcrT) SbertAttentioncs*tt||_t||_t|_dSr)r"r#rUr>routputset pruned_headsr=r@rBrCr#)s    zSbertAttention.__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)Nrrrm)lenr r>r[r^rr rarbrcrrr_union)r>headsindexrBrBrC prune_heads/s   zSbertAttention.prune_headsNFc Cs<||||||||}||d|} | f|dd} | S)Nrr)r>r) r>ryrzr{r|r}r~r self_outputsattention_outputrrBrBrCrNBs   zSbertAttention.forwardr)rOrPrQr#rrNrSrBrBr@rCr's rcr)SbertIntermediatecsDtt|j|j|_t|jt rt |j|_ dS|j|_ dSr) r"r#r$r`r'intermediate_sizer isinstance hidden_actstrrintermediate_act_fnr=r@rBrCr#]s   zSbertIntermediate.__init__cCs||}||}|Sr)rr)r>ryrBrBrCrNes  zSbertIntermediate.forwardrrBrBr@rCr[s rcr) SbertOutputcsBtt|j|j|_tj|j|jd|_t |j |_ dSr) r"r#r$r`rr'rr.r/r0r1r2r=r@rBrCr#mrzSbertOutput.__init__cCrrrrrBrBrCrNtrzSbertOutput.forwardrrBrBr@rCrkrrcs:eZdZfddZ      d ddZddZZS) SbertLayercsnt|j|_d|_t||_|j|_|j|_|jr+|js&t|dt||_ t ||_ t ||_ dS)Nrz> should be used as a decoder model if cross attention is added)r"r#chunk_size_feed_forward seq_len_dimr attentionrfadd_cross_attentionr\crossattentionr intermediaterrr=r@rBrCr#}s    zSbertLayer.__init__NFc Cs|dur |ddnd}|j|||||d} | d} |jr(| dd} | d} n| dd} d} |jro|durot|dsDtd|d|durN|d dnd}|| ||||||}|d} | |dd} |d} | | } t|j|j|j| }|f| } |jr| | f} | S) NrZrr~rrrrz'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`ro) rrfrDr\rrfeed_forward_chunkrr)r>ryrzr{r|r}r~rself_attn_past_key_valueself_attention_outputsrrpresent_key_valuecross_attn_present_key_valuecross_attn_past_key_valuecross_attention_outputs layer_outputrBrBrCrNsp        zSbertLayer.forwardcCs||}|||}|Sr)rr)r>rintermediate_outputrrBrBrCrs  zSbertLayer.feed_forward_chunkr)rOrPrQr#rNrrSrBrBr@rCr{s  Ercs8eZdZfddZ         dddZZS) SbertEncodercs:t|_tfddtjD|_d|_dS)Ncsg|]}tqSrB)r).0_r?rBrC sz)SbertEncoder.__init__..F) r"r#r?r$ ModuleListrangenum_hidden_layerslayergradient_checkpointingr=r@rrCr#s  zSbertEncoder.__init__NFTc  sb| rdnd} r dnd} r|jjrdnd} |rdnd}t|jD]o\}}| r,| |f} |dur4||nd}|dur>||nd|jre|jre|rOtdd}fdd}tj j |||||||}n ||||||}|d}|r|||df7}r| |df} |jjr| |d f} q!| r| |f} | st d d ||| | | fDSt ||| | | d S) NrBzZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...Fcsfdd}|S)Ncsg|RSrrB)inputs)modulerr~rBrCcustom_forwards zKSbertEncoder.forward..create_custom_forward..custom_forwardrB)rrr)rrCcreate_custom_forwardsz3SbertEncoder.forward..create_custom_forwardrrrrZcss|] }|dur|VqdSrrB)rvrBrBrC -s z'SbertEncoder.forward..)last_hidden_statepast_key_valuesry attentionscross_attentions) r?r enumeraterrtrainingloggerwarningr5utils checkpointtupler)r>ryrzr{r|r}r use_cacheroutput_hidden_states return_dictall_hidden_statesall_self_attentionsall_cross_attentionsnext_decoder_cachei layer_modulelayer_head_maskr layer_outputsrBrrCrNs        zSbertEncoder.forward) NNNNNNFFTrrBrBr@rCrs  rcr) SbertPoolercs*tt|j|j|_t|_dSr)r"r#r$r`r'rTanh activationr=r@rBrCr#?s zSbertPooler.__init__cCs(|dddf}||}||}|S)Nr)rr)r>ryfirst_token_tensor pooled_outputrBrBrCrNDs  zSbertPooler.forwardrrBrBr@rCr=s rcsTeZdZdZeZdZdZdgZfddZ ddZ dd d Z e fd d Z ZS)SbertPreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. bertTrc s*tj|jfi|tt||dSr)r"r# name_or_pathr )r>r?kwargsr@rBrCr#XszSbertPreTrainedModel.__init__cCst|tjr |jjjd|jjd|jdur|jj dSdSt|tj rC|jjjd|jjd|j durA|jj|j  dSdSt|tj rX|jj |jj ddSdS)zInitialize the weightsg)meanstdNg?)rr$r`weightdatanormal_r?initializer_rangebiaszero_r%rr.fill_)r>rrBrBrC _init_weights\s$      z"SbertPreTrainedModel._init_weightsFcCst|tr ||_dSdSr)rrr)r>rrcrBrBrC_set_gradient_checkpointingns  z0SbertPreTrainedModel._set_gradient_checkpointingc sj|dd}|dd}t||fi|}|dur&tdi|}||}|Stt|jdd|i|}|S)alInstantiate the model. Args: kwargs: Input args. model_dir: The model dir used to load the checkpoint and the label information. num_labels: An optional arg to tell the model how many classes to initialize. Method will call utils.parse_label_mapping if num_labels is not input. label2id: An optional label2id mapping, which will cover the label2id in configuration (if exists). Returns: The loaded model, which is initialized by transformers.PreTrainedModel.from_pretrained model_dirNcfgpretrained_model_name_or_pathrB)poprrr"r from_pretrained)clsrrr model_argsr?modelr@rBrC _instantiaters  z!SbertPreTrainedModel._instantiate)F)rOrPrQrRr config_classbase_model_prefixsupports_gradient_checkpointing_keys_to_ignore_on_load_missingr#rr classmethodrrSrBrBr@rCrMs  rc@sBeZdZUdZejed<dZee e ejfed<dZ e ed<dS))AttentionBackboneModelOutputWithEmbeddingNembedding_outputlogitsr) rOrPrQr r5 FloatTensor__annotations__r rrrrdictrBrBrBrCr s r ) module_namecsdeZdZdZddeffdd ZddZdd Zd d Z dd dZ Z S) SbertModela+The StructBERT Model transformer outputting raw hidden-states without any specific head on top. This model inherits from :class:`~transformers.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 `__ 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 (:class:`~modelscope.models.nlp.structbert.SbertConfig`): 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 :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights. 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 `__ 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 :obj:`is_decoder` argument of the configuration set to :obj:`True`. To be used in a Seq2Seq model, the model needs to initialized with both :obj:`is_decoder` argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an input to the forward pass. Tr?c sDt|||_t||_t||_|rt|nd|_| dSr) r"r#r?rrMrencoderrpooler init_weights)r>r?add_pooling_layerrr@rBrCr#s    zSbertModel.__init__cCs|jjSrrMr))r>rBrBrCget_input_embeddingsszSbertModel.get_input_embeddingscCs ||j_dSrr)r>rcrBrBrCset_input_embeddingss zSbertModel.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)r>heads_to_prunerrrBrBrC _prune_headsszSbertModel._prune_headsNc! KsT| dur| n|jj} | dur| n|jj} | dur| n|jj} |jjr-| dur(| n|jj} nd} |dur;|dur;td|durD|}n|durQ|dd}ntd|\}}|dur`|jn|j}| durp| ddj dnd}|durt j |||f|d}|durt |j d r|j jddd|f}|||}|}n t j|t j|d }||||}|jjr|dur|\}}}||f}|durt j ||d}||}nd}|||jj}|j |||||d d \}}|j|||||| | | | | d }|d}|jdur ||nd} | s|| f|dd|fSt|| |j|j|j|j|dS)u Args: input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using :class:`~modelscope.models.nlp.structbert.SbertTokenizer`. See :meth:`transformers.PreTrainedTokenizer.encode` and :meth:`transformers.PreTrainedTokenizer.__call__` for details. attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `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**. token_type_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, 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. position_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, 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]``. head_mask (:obj:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(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 (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`): Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert :obj:`input_ids` indices into associated vectors than the model's internal embedding lookup matrix. output_attentions (:obj:`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 (:obj:`bool`, `optional`): Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for more detail. return_dict (:obj:`bool`, `optional`): Whether or not to return a :class:`~transformers.ModelOutput` instead of a plain tuple. encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(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 (:obj:`torch.FloatTensor` of shape :obj:`(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 (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(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 :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids` (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)` instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`. use_cache (:obj:`bool`, `optional`): If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up decoding (see :obj:`past_key_values`). Returns: Returns `modelscope.outputs.AttentionBackboneModelOutputWithEmbedding` Examples: >>> from modelscope.models import Model >>> from modelscope.preprocessors import Preprocessor >>> model = Model.from_pretrained('damo/nlp_structbert_backbone_base_std', task='backbone') >>> preprocessor = Preprocessor.from_pretrained('damo/nlp_structbert_backbone_base_std') >>> print(model(**preprocessor('这是个测试'))) NFzDYou cannot specify both input_ids and inputs_embeds at the same timerz5You have to specify either input_ids or inputs_embedsrrZ)r rrT)rErrrFrGrH) rzr{r|r}rrrrrr)r pooler_outputrryrrr )r?rruse_return_dictrfrr\r;r shaper5onesrDrMrr7r:r<get_extended_attention_maskinvert_attention_mask get_head_maskrrrr rryrr)!r>rErzrrr{rFr|r}rrrrrrrI batch_sizerJr rGrKrLextended_attention_maskencoder_batch_sizeencoder_sequence_lengthrencoder_hidden_shapeencoder_extended_attention_maskr original_embedsencoder_outputssequence_outputrrBrBrCrNs^        zSbertModel.forward)T) NNNNNNNNNNNNN) rOrPrQrRrr#rrrrNrSrBrBr@rCrs(  r)6rRru dataclassesrtypingrrr5torch.nnr$torch.utils.checkpoint packagingrtransformers.activationsrtransformers.modeling_utilsrrr r modelscope.metainfor modelscope.modelsr r modelscope.models.builderrmodelscope.outputsrmodelscope.utils.constantrmodelscope.utils.loggerrmodelscope.utils.nlp.utilsr configurationrrModulerrUrrrrrrrrr register_modulebackbone structbertrrBrBrBrCsD           W4]eA