o h@sdZddlZddlmZddlmZmZmZddlZddl Zddlm Z ddl m Z m Z mZddlmZdd lmZmZmZmZmZmZdd lmZdd lmZmZmZdd lmZm Z m!Z!m"Z"d dl#m$Z$e"%e&Z'dZ(dZ)dgZ*ddZ+ddZ,ddZ-ddZ.Gdddej/j0Z1Gdddej/j0Z2Gddde j3Z4Gdd d e j3Z5Gd!d"d"e j3Z6Gd#d$d$e j3Z7Gd%d&d&e j3Z8Gd'd(d(e j3Z9Gd)d*d*e j3Z:Gd+d,d,e j3Z;Gd-d.d.e j3ZGd3d4d4eZ?d5Z@d6ZAe d7e@Gd8d9d9e?ZBe d:e@Gd;d<dd>e j3ZDe d?e@Gd@dAdAe?ZEe dBe@GdCdDdDe?ZFe dEe@GdFdGdGe?ZGe dHe@GdIdJdJe?ZHdS)Kz PyTorch YOSO model.N)Path)OptionalTupleUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)"BaseModelOutputWithCrossAttentionsMaskedLMOutputMultipleChoiceModelOutputQuestionAnsweringModelOutputSequenceClassifierOutputTokenClassifierOutput)PreTrainedModel)apply_chunking_to_forward find_pruneable_heads_and_indicesprune_linear_layer)add_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardlogging) YosoConfigzuw-madison/yoso-4096rcCsVzddlm}dd}|gd}|d|ddddlaWdSty*daYd Sw) Nr)loadcs,ttjjjddfdd|DS)Nkernelsyosocsg|]}|qSr).0file src_folderr\/var/www/html/ai/venv/lib/python3.10/site-packages/transformers/models/yoso/modeling_yoso.py <z:load_cuda_kernels..append_root..)r__file__resolveparent)filesrr"r$ append_root:sz&load_cuda_kernels..append_root)zfast_lsh_cumulation_torch.cppzfast_lsh_cumulation.cuzfast_lsh_cumulation_cuda.cufast_lsh_cumulationT)verboseF)torch.utils.cpp_extensionrr,lsh_cumulation Exception)rr+ src_filesrrr$load_cuda_kernels5s  r2cCsJt|trg}|D]}|s|}||q |S|s#|}|SN) isinstancelist is_contiguous contiguousappend input_tensorsouttensorrrr$ to_contiguousLs  r=cCsFt|trg}|D]}|tjj|dddq |Stjj|dddS)N)pdim)r4r5r8r functional normalizer9rrr$rCZs rCc Cst|dkr tdt|dkrtdtj|d|d|||jd}dtj||jd}t|||d|d||}t|||d|d||}|dk }|dk } tj ||dd } tj | |dd } | | fS) Nr zQuery has incorrect size.zKey has incorrect size.rr>devicerr?rA) lensize ValueErrortorchrandnrEarangematmulreshapeintsum) querykeynum_hashhash_lenrmat raise_powquery_projectionkey_projection query_binary key_binary query_hashrrr$hashingds$$$  r\c@$eZdZeddZeddZdS)YosoCumulationc Cs|d}dtt||ddtj|}||dddddf|dddddf}t||} |||||||||_| S)N hash_code_lenrr?)rJacosrM transposemathpisave_for_backwardconfig) ctx query_maskkey_maskrQrRvaluerfr_ expectationcumulation_valuerrr$forwardxs(0 zYosoCumulation.forwardcCst|}|j\}}}}}}|j}|d} t||dd|} t| | d|} t| dd| d|} t|dd|} dd| | | dfS)Nr_r?r`r>)r= saved_tensorsrfrJrMrb)rggradrhrirkrQrRrjrfr_ weighted_exp grad_querygrad_key grad_valuerrr$backwardszYosoCumulation.backwardN__name__ __module__ __qualname__ staticmethodrmrtrrrr$r^ws   r^c@r])YosoLSHCumulationc CsV|d|dkrtd|d|dkrtd|d|dkr*td|d|dkr8td|d|dkrFtd|d|dkrTtd t|||||g\}}}}}|j}|d }|d } td| } |d rt|||||| |d\} } n t|||| \} } t|| || || |d} |||| | |||||_ | S) Nrz6Query mask and Key mask differ in sizes in dimension 0z3Query mask and Query differ in sizes in dimension 0z1Query mask and Key differ in sizes in dimension 0z8Query mask and Value mask differ in sizes in dimension 0rz,Key and Value differ in sizes in dimension 1r>z,Query and Key differ in sizes in dimension 2rSr_ use_fast_hash) rHrIr=is_cudarOr/ fast_hashr\rerf)rgrhrirQrRrjrfuse_cudarSr_hashtable_capacityquery_hash_code key_hash_coderlrrr$rms8  zYosoLSHCumulation.forwardc Csjt|}|j\}}}}}}}|j} |j} | d} td| } | drSt|||||| | d} t||||||| d|| | d }t||||||| d|| | d }nZdtt || ddt j | }||dddddf|dddddf}t || dd|}t || d|}t | dd| d|}t | dd|} dd||| dfS)Nr_r> lsh_backwardrr?r`) r=rnrfr|rOr/lsh_weighted_cumulationrJrarMrbrcrd)rgrorhrirrrQrRrjrfr~r_rrsrqrrrkrprrr$rtsR    ( 0zYosoLSHCumulation.backwardNrurrrr$rzs  %rzcs*eZdZdZfddZdddZZS)YosoEmbeddingszGConstruct the embeddings from word, position and token_type embeddings.csttj|j|j|jd|_t|jd|j|_ t|j |j|_ tj |j|j d|_ t|j|_|jdt|jddddt|dd |_|jd tj|jtj|jjd dddS) N) padding_idxr>eps position_ids)rr?F) persistentposition_embedding_typeabsolutetoken_type_idsdtyperE)super__init__r Embedding vocab_size hidden_size pad_token_idword_embeddingsmax_position_embeddingsposition_embeddingstype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsDropouthidden_dropout_probdropoutregister_bufferrJrLexpandgetattrrzerosrrHlongrEselfrf __class__rr$rs  zYosoEmbeddings.__init__Nc Cs|dur |}n|dd}|d}|dur$|jddd|f}|durNt|drC|jddd|f}||d|}|}n tj|tj|jjd}|durW| |}| |} || } |j dkrn| |} | | 7} | | } || } | S)Nr?rrrrr)rHrhasattrrrrJrrrErrrrrr) r input_idsrr inputs_embeds input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expandedr embeddingsrrrr$rm s,        zYosoEmbeddings.forward)NNNNrvrwrx__doc__rrm __classcell__rrrr$rs rc0eZdZd fdd ZddZd ddZZS) YosoSelfAttentionNcsJt|j|jdkrt|dstd|jd|jd|j|_t|j|j|_|j|j|_t |j|j|_ t |j|j|_ t |j|j|_ t |j|_|dur[|n|j|_|j|_|j|_|jdu|_|j|_|j|_|j|_|j|j|j|jd|_|jdurt j|j|j|jdf|jddfd |jd |_dSdS) Nrembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads ())r_r{rSrrr>F) in_channels out_channels kernel_sizepaddingbiasgroups)rrrnum_attention_headsrrIrOattention_head_size all_head_sizerLinearrQrRrjrattention_probs_dropout_probrruse_expectationr_ conv_windowuse_convr{rSr lsh_configConv2dconvrrfrrrr$r-sJ      zYosoSelfAttention.__init__cCs6|dd|j|jf}|j|}|ddddS)Nr?rr>rr )rHrrviewpermute)rlayernew_layer_shaperrr$transpose_for_scoresZs z&YosoSelfAttention.transpose_for_scoresFcCs>||}|||}|||}||}|jr.|||ddddddf}|\} } } } || | | | }|| | | | }|| | | | }d|d}| d| d| | |  }d} |j s| | kr| | | | | f}t j |t j||jdgdd}t j |t j||jdgdd}t j |t j||jdgdd}|j s|jrt||g\}}|j rt||||||j}n t||||||j}|j s| | kr|ddddd| f}t|}|| | | | }|jr||7}|dd dd }|dd |jf}|j|}|r||f}|S|f}|S) N?g@r rDr?rFrr>r r`)rQrrRrjrrrHrNsqueezerepeatrOrrJcatrrEtrainingrCr^applyrrzrr7rr)r hidden_statesattention_maskoutput_attentionsmixed_query_layer key_layer value_layer query_layerconv_value_layer batch_size num_headsseq_lenhead_dim gpu_warp_sizepad_size context_layernew_context_layer_shapeoutputsrrr$rm_sp  "    zYosoSelfAttention.forwardr3NF)rvrwrxrrrmrrrrr$r,s-rc8eZdZfddZdejdejdejfddZZS)YosoSelfOutputcsBtt|j|j|_tj|j|jd|_t|j |_ dSNr) rrrrrdenserrrrrrrrr$r zYosoSelfOutput.__init__r input_tensorreturncC&||}||}|||}|Sr3rrrrrrrrr$rm  zYosoSelfOutput.forwardrvrwrxrrJTensorrmrrrrr$r $rcr) YosoAttentionNcs.tt||d|_t||_t|_dS)N)r)rrrrroutputset pruned_headsrrrr$rs   zYosoAttention.__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)NrrrF)rGrrrrrrrQrRrjrrrunion)rheadsindexrrr$ prune_headss zYosoAttention.prune_headsFcCs4||||}||d|}|f|dd}|S)Nrr)rr)rrrr self_outputsattention_outputrrrr$rmszYosoAttention.forwardr3r)rvrwrxrrrmrrrrr$rsrc2eZdZfddZdejdejfddZZS)YosoIntermediatecsDtt|j|j|_t|jt rt |j|_ dS|j|_ dSr3) rrrrrintermediate_sizerr4 hidden_actstrr intermediate_act_fnrrrr$rs   zYosoIntermediate.__init__rrcC||}||}|Sr3)rrrrrrr$rm  zYosoIntermediate.forwardrrrrr$rs rcr) YosoOutputcsBtt|j|j|_tj|j|jd|_t |j |_ dSr) rrrrrrrrrrrrrrrr$rrzYosoOutput.__init__rrrcCrr3rrrrr$rmrzYosoOutput.forwardrrrrr$r rr cs.eZdZfddZd ddZddZZS) YosoLayercsBt|j|_d|_t||_|j|_t||_t ||_ dSNr) rrchunk_size_feed_forward seq_len_dimr attentionadd_cross_attentionr intermediater rrrrr$rs   zYosoLayer.__init__NFcCsF|j|||d}|d}|dd}t|j|j|j|}|f|}|S)N)rrr)rrfeed_forward_chunkr r )rrrrself_attention_outputsrr layer_outputrrr$rm s  zYosoLayer.forwardcCs||}|||}|Sr3)rr)rrintermediate_outputrrrr$rs  zYosoLayer.feed_forward_chunkr)rvrwrxrrmrrrrrr$r s   r cs0eZdZfddZ     dddZZS) YosoEncodercs:t|_tfddtjD|_d|_dS)Ncsg|]}tqSr)r )r _rfrr$r% r&z(YosoEncoder.__init__..F) rrrfr ModuleListrangenum_hidden_layersrgradient_checkpointingrrrr$rs   zYosoEncoder.__init__NFTc Cs|rdnd}|r dnd}t|jD].\} } |r||f}|jr,|jr,|| j|||} n| |||} | d}|r?|| df}q|rG||f}|sUtdd|||fDSt|||dS)Nrrrcss|] }|dur|VqdSr3r)r vrrr$ Esz&YosoEncoder.forward..)last_hidden_stater attentions) enumeraterrr_gradient_checkpointing_func__call__tupler ) rrr head_maskroutput_hidden_states return_dictall_hidden_statesall_self_attentionsi layer_module layer_outputsrrr$rm#s4     zYosoEncoder.forward)NNFFTrvrwrxrrmrrrrr$rs  rcr)YosoPredictionHeadTransformcsVtt|j|j|_t|jtrt |j|_ n|j|_ tj |j|j d|_ dSr) rrrrrrr4rrr transform_act_fnrrrrrr$rOs  z$YosoPredictionHeadTransform.__init__rrcCs"||}||}||}|Sr3)rr.rrrrr$rmXs   z#YosoPredictionHeadTransform.forwardrrrrr$r-Ns  r-cs$eZdZfddZddZZS)YosoLMPredictionHeadcsLtt||_tj|j|jdd|_t t |j|_ |j |j_ dS)NF)r) rrr- transformrrrrdecoder ParameterrJrrrrrr$ras  zYosoLMPredictionHead.__init__cCrr3)r0r1rrrr$rmnrzYosoLMPredictionHead.forwardr,rrrr$r/`s  r/cr)YosoOnlyMLMHeadcstt||_dSr3)rrr/ predictionsrrrr$rvs zYosoOnlyMLMHead.__init__sequence_outputrcCs||}|Sr3)r4)rr5prediction_scoresrrr$rmzs zYosoOnlyMLMHead.forwardrrrrr$r3us r3c@s$eZdZdZeZdZdZddZdS)YosoPreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. rTcCst|tjr |jjjd|jjd|jdur|jj dSdSt|tj rC|jjjd|jjd|j durA|jj|j  dSdSt|tj rX|jj |jj ddSdS)zInitialize the weightsg)meanstdNr)r4rrweightdatanormal_rfinitializer_rangerzero_rrrfill_)rmodulerrr$ _init_weightss      z!YosoPreTrainedModel._init_weightsN) rvrwrxrr config_classbase_model_prefixsupports_gradient_checkpointingrArrrr$r7s  r7aG This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`YosoConfig`]): 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. a5 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) 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 YOSO Model transformer outputting raw hidden-states without any specific head on top.cseZdZfddZddZddZddZee d e e e e d  dd eejd eejdeejdeejdeejdeejdeedeedeedeee ffddZZS) YosoModelcs2t|||_t||_t||_|dSr3)rrrfrrrencoder post_initrrrr$rs    zYosoModel.__init__cCs|jjSr3rrrrrr$get_input_embeddingsszYosoModel.get_input_embeddingscCs ||j_dSr3rH)rrjrrr$set_input_embeddingss zYosoModel.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)itemsrFrrr)rheads_to_prunerrrrr$ _prune_headsszYosoModel._prune_headsbatch_size, sequence_length checkpoint output_typerBNrrrrr$rrr%r&rc Cs|dur|n|jj}|dur|n|jj}| dur| n|jj} |dur*|dur*td|dur9||||} n|durF|dd} ntd| \} } |durU|jn|j} |duretj | | f| d}|durt |j dr|j j ddd| f}| | | }|}n tj| tj| d}||| }|||jj}|j ||||d}|j|||||| d}|d }| s|f|d dSt||j|j|jd S) NzDYou cannot specify both input_ids and inputs_embeds at the same timer?z5You have to specify either input_ids or inputs_embedsrDrr)rrrr)rr$rr%r&rr)rrrcross_attentions)rfrr%use_return_dictrI%warn_if_padding_and_no_attention_maskrHrErJonesrrrrrrget_extended_attention_mask get_head_maskrrFr rrrS)rrrrrr$rrr%r&rrrrErrextended_attention_maskembedding_outputencoder_outputsr5rrr$rms^     zYosoModel.forward) NNNNNNNNN)rvrwrxrrJrKrNrYOSO_INPUTS_DOCSTRINGformatr_CHECKPOINT_FOR_DOCr _CONFIG_FOR_DOCrrJrboolrrrmrrrrr$rEsR      rEz2YOSO Model with a `language modeling` head on top.cseZdZddgZfddZddZddZee d e e e e d  dd eejd eejdeejdeejdeejdeejdeejdeedeedeedeee ffddZZS)YosoForMaskedLMzcls.predictions.decoder.weightzcls.predictions.decoder.biascs,t|t||_t||_|dSr3)rrrErr3clsrGrrrr$rNs    zYosoForMaskedLM.__init__cCs |jjjSr3rbr4r1rIrrr$get_output_embeddingsWs z%YosoForMaskedLM.get_output_embeddingscCs||jj_dSr3rc)rnew_embeddingsrrr$set_output_embeddingsZsz%YosoForMaskedLM.set_output_embeddingsrOrPNrrrrr$rlabelsrr%r&rc  Cs| dur| n|jj} |j|||||||| | d } | d} || } d}|dur8t}|| d|jj|d}| sN| f| dd}|durL|f|S|St|| | j| j dS)a labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. Nrrrr$rrr%r&rr?rlosslogitsrr) rfrTrrbrrrr rr)rrrrrr$rrgrr%r&rr5r6masked_lm_lossloss_fctrrrr$rm]s6 zYosoForMaskedLM.forward NNNNNNNNNN)rvrwrx_tied_weights_keysrrdrfrr\r]rr^r r_rrJrr`rrrmrrrrr$raJsX       racs(eZdZdZfddZddZZS)YosoClassificationHeadz-Head for sentence-level classification tasks.csFtt|j|j|_t|j|_t|j|j |_ ||_ dSr3) rrrrrrrrr num_labelsout_projrfrrrr$rs  zYosoClassificationHead.__init__cKsR|dddddf}||}||}t|jj|}||}||}|S)Nr)rrr rfrrr)rfeatureskwargsxrrr$rms    zYosoClassificationHead.forwardrrrrr$rps rpzYOSO Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks.ceZdZfddZeedeee e d          dde e j de e j de e j d e e j d e e j d e e j d e e j d e ede ede edeee ffddZZS)YosoForSequenceClassificationcs4t||j|_t||_t||_|dSr3)rrrqrErrp classifierrGrrrr$rs    z&YosoForSequenceClassification.__init__rOrPNrrrrr$rrgrr%r&rc  Csh| dur| n|jj} |j|||||||| | d } | d} || } d}|dur|jjdurQ|jdkr7d|j_n|jdkrM|jtjksH|jtj krMd|j_nd|j_|jjdkrot }|jdkri|| | }n+|| |}n%|jjdkrt }|| d|j| d}n|jjdkrt}|| |}| s| f| dd}|dur|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). Nrhrr regressionsingle_label_classificationmulti_label_classificationr?ri)rfrTrrx problem_typerqrrJrrOr rrrrrrr)rrrrrr$rrgrr%r&rr5rkrjrmrrrr$rmsT    "       z%YosoForSequenceClassification.forwardrn)rvrwrxrrr\r]rr^rr_rrJrr`rrrmrrrrr$rwsR      rwzYOSO Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks.crv)YosoForMultipleChoicecsDt|t||_t|j|j|_t|jd|_| dSr ) rrrErrrrpre_classifierrxrGrrrr$rs   zYosoForMultipleChoice.__init__z(batch_size, num_choices, sequence_lengthrPNrrrrr$rrgrr%r&rc  Cs| dur| n|jj} |dur|jdn|jd} |dur%|d|dnd}|dur4|d|dnd}|durC|d|dnd}|durR|d|dnd}|dure|d|d|dnd}|j|||||||| | d } | d} | dddf}||}t|}| |}|d| }d}|durt }|||}| s|f| dd}|dur|f|S|St ||| j | j dS)aJ labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See `input_ids` above) Nrr?r`rhrri)rfrTshaperrHrr~rReLUrxrrrr)rrrrrr$rrgrr%r& num_choicesr hidden_state pooled_outputrkreshaped_logitsrjrmrrrr$rmsP      zYosoForMultipleChoice.forwardrn)rvrwrxrrr\r]rr^rr_rrJrr`rrrmrrrrr$r}sR      r}zYOSO 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.crv)YosoForTokenClassificationcsJt||j|_t||_t|j|_t |j |j|_ | dSr3) rrrqrErrrrrrrrxrGrrrr$ris   z#YosoForTokenClassification.__init__rOrPNrrrrr$rrgrr%r&rc  Cs | dur| n|jj} |j|||||||| | d } | d} || } || } d}|duret}|durX|ddk}| d|j}t ||dt |j  |}|||}n || d|j|d}| s{| f| dd}|dury|f|S|St || | j| jdS)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]`. Nrhrr?rri)rfrTrrrxrrrqrJwherer< ignore_indextype_asrrr)rrrrrr$rrgrr%r&rr5rkrjrm active_loss active_logits active_labelsrrrr$rmtsF   z"YosoForTokenClassification.forwardrn)rvrwrxrrr\r]rr^rr_rrJrr`rrrmrrrrr$rcsR      rzYOSO Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute `span start logits` and `span end logits`).cseZdZfddZeedeee e d           dde e j de e j de e j d e e j d e e j d e e j d e e j d e e j de ede ede edeee ffddZZS)YosoForQuestionAnsweringcsBt|d|_|j|_t||_t|j|j|_| dS)Nr>) rrrqrErrrr qa_outputsrGrrrr$rs   z!YosoForQuestionAnswering.__init__rOrPNrrrrr$rstart_positions end_positionsrr%r&rc  Cs@| dur| n|jj} |j||||||| | | d } | d} || }|jddd\}}|d}|d}d}|dur~|dur~t|dkrK|d}t|dkrX|d}|d}|d|}|d|}t |d}|||}|||}||d}| s||f| dd}|dur|f|S|St |||| j | j d S) a start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. Nrhrrr?rF)rr>)rj start_logits end_logitsrr) rfrTrrsplitrrGrHclamprrrr)rrrrrr$rrrrr%r&rr5rkrr total_loss ignored_indexrm start_lossend_lossrrrr$rmsP            z YosoForQuestionAnswering.forward) NNNNNNNNNNN)rvrwrxrrr\r]rr^rr_rrJrr`rrrmrrrrr$rsX       r)IrrcpathlibrtypingrrrrJtorch.utils.checkpointrtorch.nnrrr activationsr modeling_outputsr r rrrrmodeling_utilsr pytorch_utilsrrrutilsrrrrconfiguration_yosor get_loggerrvloggerr^r_"YOSO_PRETRAINED_MODEL_ARCHIVE_LISTr2r=rCr\autogradFunctionr^rzModulerrrrrr r rr-r/r3r7YOSO_START_DOCSTRINGr\rErarprwr}rrrrrr$s        Z<!2  2oMVVP