sg`dZddlmZddlmZddlmZmZmZddl Z ddl Z ddl mZddlmZmZmZdd lmZmZmZmZmZmZmZmZmZmZmZdd lm Z m!Z!m"Z"dd l#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z)d d l*m+Z+e(jXe-Z.dZ/dZ0GddejbjdZ3GddejbjdZ4GddejbjdZ5eGddejbjdZ6GddeZ7eGdde$Z8dZ9dZ:e&de9Gdd e7Z;e&d!e9Gd"d#e7eZ<e&d$e9Gd%d&e7Z=e&d'e9Gd(d)e7eZ>y)*zTF 2.0 OpenAI GPT model.) annotations) dataclass)OptionalTupleUnionN)get_tf_activation)TFBaseModelOutputTFCausalLMOutputTFSequenceClassifierOutput) TFCausalLanguageModelingLossTFConv1DTFModelInputTypeTFPreTrainedModelTFSequenceClassificationLossTFSequenceSummaryTFSharedEmbeddingsget_initializerkeraskeras_serializable unpack_inputs)check_embeddings_within_bounds shape_liststable_softmax) ModelOutputadd_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardloggingreplace_return_docstrings)OpenAIGPTConfigzopenai-community/openai-gptr"cZeZdZd fd ZdZedZd dZdZdZ d dZ d dZ xZ S) TFAttentionc <t|di||}||jzdk(sJd|d|j|j|_||_||_|j |_t |dz||jd|_t |||jd|_ tjj|j|_tjj|j|_||_t%|_y) NrzHidden dimension z" not dividable by number of heads rc_attninitializer_rangenamec_proj)super__init__n_head split_sizescaleoutput_attentionsrr(r&r*rlayersDropout attn_pdrop attn_dropout resid_pdrop resid_dropoutn_stateset pruned_heads)selfnxconfigr0kwargsr8 __class__s `/var/www/html/venv/lib/python3.12/site-packages/transformers/models/openai/modeling_tf_openai.pyr-zTFAttention.__init__<s "6" fmm #q ( Z wi'I&-- Y Z (mm ! !'!9!9w{B&BZBZaij wf>V>V]ef !LL001B1BC"\\11&2D2DE EcyNr+)r;headss r@ prune_headszTFAttention.prune_headsPs rAc~tj|dddf}tj|}|||z |zk\}|S)z 1's in the lower triangle, counting from the lower right corner. Same as tf.matrix_band_part(tf.ones([nd, ns]), -1, ns-nd), but doesn't produce garbage on TPUs. N)tfrange)ndnsijms r@causal_attention_maskz!TFAttention.causal_attention_maskSs> HHRLD ! HHRL R" rActj||d}|jrOtjt |d|j } |tj j| z }t |\} } } } tj|j| | |j } tj| dd| | g} || zdd| z zz }|&tj||j }||z}t|d}|j||}|||z}tj||g}|r|j||S) NT) transpose_bdtyper!g@axistraining) rGmatmulr0castrrSmathsqrtrNreshaperr5append)r;qkvattention_mask head_maskr1rWwdk_rIrJboutputss r@_attnzTFAttention._attn^s- IIa - ::Ar*!'':BBGGLL$$A"!} 1b" GGD..r26agg F JJq1aR. ) EC1q5M !  %WW^177CNN"A 12 &   a(  3  I A99Q?#  NN1 rActj|gd}t|}|dd|d|dzgz}tj||S)Nrr!rrQ)rG transposerr\r;xx_shape new_x_shapes r@ merge_headszTFAttention.merge_heads|sL LLL )Q-crlgbkGBK&?%@@ zz![))rAct|}|dd|j|d|jzgz}tj||}tj|dS)NrQrj)rr.rGr\rmrns r@ split_headszTFAttention.split_headssRQ-crldkk72;$++3M%NN JJq+ &||A|,,rAc ||j|}tj|dd\}}}|j|}|j|}|j|}|j |||||||} | d} |j | } |j | } |j| |} | g| ddz} | S)NrrkrTrVrr!)r&rGsplitrtrhrrr*r7) r;rorarbr1rWquerykeyvalue attn_outputsargs r@callzTFAttention.calls KKNHHQ2sE  's#  'zz%e^YPaltzu O   Q  KKN   q8  4# QR((rAc|jryd|_t|dd]tj|jj 5|jj dd|jdzgdddt|dd[tj|jj 5|jj dd|jgdddyy#1swYrxYw#1swYyxYw)NTr&rr*) builtgetattrrG name_scoper&r)buildr8r*r; input_shapes r@rzTFAttention.builds ::  44 ( 4t{{//0 B !!4t||a/?"@A B 44 ( 4t{{//0 > !!4t||"<= > > 5 B B > >s ,C(5)C4(C14C=FrC) __name__ __module__ __qualname__r-rE staticmethodrNrhrrrtr|r __classcell__r?s@r@r$r$;s:"( <* - " >rAr$c.eZdZfdZddZddZxZS)TFMLPc Lt|di||j}t|||jd|_t|||jd|_td|_tjj|j|_ ||_||_y)Nc_fcr'r*gelur+)r,r-n_embdrr(rr*r actrr2r3r6dropoutr<r8)r;r8r=r>r<r?s r@r-zTFMLP.__init__s "6" ]]WbFV>V]ef $V,||++F,>,>?  rAc|j|j|}|j|}|j||}|S)NrV)rrr*r)r;rorWhh2s r@r|z TFMLP.calls< HHTYYq\ " [[^ \\"x\ 0 rAc|jryd|_t|ddZtj|jj 5|jj dd|jgdddt|dd[tj|jj 5|jj dd|jgdddyy#1swYrxYw#1swYyxYw)NTrr*) r~rrGrrr)rr8r*r<rs r@rz TFMLP.builds ::  4 & 2tyy~~. < tT\\ :; < 44 ( 4t{{//0 9 !!4tww"78 9 9 5 < < 9 9s )C%2)C1%C.1C:rrCrrrr-r|rrrs@r@rrs 9rArc0eZdZdfd ZddZddZxZS)TFBlockc ^t|di||j}t|||d|_t j j|jd|_ td|z|d|_ t j j|jd|_ ||_ y) Nattnr)ln_1)epsilonr)mlpln_2r+)r,r-rr$rrr2LayerNormalizationlayer_norm_epsilonrrrrr<)r;r=r0r>r<r?s r@r-zTFBlock.__init__s "6" ]]FE? LL33F9>OZbic N IIa!e  HHQH * IIa!e # AB'rAc|jryd|_t|ddMtj|jj 5|jj ddddt|ddZtj|jj 5|jj dd|jgdddt|ddMtj|jj 5|jj ddddt|dd[tj|jj 5|jj dd|jgdddyy#1swY4xYw#1swYxYw#1swYxYw#1swYyxYw)NTrrrr) r~rrGrrr)rrr<rrrs r@rz TFBlock.buildsS ::  4 & 2tyy~~. & % & 4 & 2tyy~~. 7 tTWW 56 7 4 % 1txx}}- %t$ % 4 & 2tyy~~. 7 tTWW 56 7 7 3 & & 7 7 % % 7 7s0 F%)F& F2&)F>F#&F/2F;>GrrCrrs@r@rrs 7rArceZdZeZfdZddZdZdZdZ e d d dZ xZ S) TFOpenAIGPTMainLayerc Xt||i|||_|j|_|j|_|j |_|j|_|j|_ |j|_ |j|_ t|j|j|jd|_tj j#|j$|_t)|jDcgc]}t+|dd|c}|_ycc}w)N tokens_embedr'Tzh_._)r0r))r,r-r=output_hidden_statesr1use_return_dict return_dictn_layernum_hidden_layersr n_positionsr(r vocab_sizerrr2r3 embd_pdropdroprHrr)r;r=inputsr>rKr?s r@r-zTFOpenAIGPTMainLayer.__init__s &+F+ $*$?$?!!'!9!9!11!'mm !--!'!9!9.   v}}@X@X_m LL(():):; HMfnnH]^1'&T!:>^^sD'ctjd5|jd|j|jgt |j |_ddd|jryd|_t|ddMtj|jj5|jjddddt|ddK|jD];}tj|j5|jdddd=yy#1swYxYw#1swYnxYw#1swYaxYw)Npositions_embed embeddings)r)shape initializerTrr)rGr add_weightrrrr(rr~rrr)rr)r;rlayers r@rzTFOpenAIGPTMainLayer.builds ]], - #'??!''5+D,B,BC$3$D   ::  4 . :t00556 .!!''- . 4d # / &]]5::.&KK%&& & 0   . .&&s$AD(+D4 E(D14D=E c|jSrC)rr;s r@get_input_embeddingsz)TFOpenAIGPTMainLayer.get_input_embeddingss   rAc`||j_t|d|j_y)Nr)rweightrrr;rys r@set_input_embeddingsz)TFOpenAIGPTMainLayer.set_input_embeddingss(#( '1%'8';$rAct)zv Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} )NotImplementedError)r;heads_to_prunes r@ _prune_headsz!TFOpenAIGPTMainLayer._prune_headss "!rAc $| | td|'t|} tj|d| dg}n|t|dd} n td|-tjtj | dd}|tj|| ddd| df}tj d} tj|| j}tjtj| |tj d }nd}|tdg|jz}tj|dt|dg}|3t||jj|j!|d }tj"|j$|} |Ytj|dt|dg}t||jjd |j!|d }nd}|| z|z}|j'|| }| t|dgz}|rdnd}|rdnd}t)|j*D]B\}}|r|tj||fz}||||||| }|d}|s:||dfz}Dtj||}|r||fz}|r/| dddgzt|dddzt-fd|D}| st-d|||fDSt/|||S)NzDYou cannot specify both input_ids and inputs_embeds at the same timerQz5You have to specify either input_ids or inputs_embedsrrTr!g?rRg embeddingmodetoken_type_idsrVr+rlc3JK|]}tj|ywrC)rGr\).0tattention_output_shapes r@ z,TFOpenAIGPTMainLayer.call..s"aQ2::a1G#H"as #c3&K|] }|| ywrCr+)rr`s r@rz,TFOpenAIGPTMainLayer.call..shqZ[Zghs)last_hidden_state hidden_states attentions) ValueErrorrrGr\ expand_dimsrHconstantrYrSmultiplysubtractrrrr=rrgatherrr enumeratertupler )r; input_idsrar position_idsrb inputs_embedsr1rrrWrone_cstposition_embedstoken_type_embedsr output_shapeall_attentionsall_hidden_statesrKblockrgrs @r@r|zTFOpenAIGPTMainLayer.calls:  ]%>cd d  "$Y/K 9r;r?.CDI  &$]3CR8KTU U  >>"((;r?*C!LL  %  ZZQA{[\~8^_Nkk#&GWW^7==IN[[Wn)Mr{{[cOdeN!N  % %!7!77Izz,Z 5Mb5Q0RS  *9dkk6L6L M --ik-JM))D$8$8,G  %ZZZ=WXZ=[8\]N *>4;;;Q;QSc d $ 1 1.{ 1 S  ! %7:KK  -( C "j&?&C%DD 0d"6BD!$&&) @HAu#$5MS_9`8b$b!! !! G$AJM !/71:-!? @ =,?  1]4D D  %0"%5% TensorFlow models and layers in `transformers` accept two formats as input: - having all inputs as keyword arguments (like PyTorch models), or - having all inputs as a list, tuple or dict in the first positional argument. The reason the second format is supported is that Keras methods prefer this format when passing inputs to models and layers. Because of this support, when using methods like `model.fit()` things should "just work" for you - just pass your inputs and labels in any format that `model.fit()` supports! If, however, you want to use the second format outside of Keras methods like `fit()` and `predict()`, such as when creating your own layers or models with the Keras `Functional` API, there are three possibilities you can use to gather all the input Tensors in the first positional argument: - a single Tensor with `input_ids` only and nothing else: `model(input_ids)` - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: `model([input_ids, attention_mask])` or `model([input_ids, attention_mask, token_type_ids])` - a dictionary with one or several input Tensors associated to the input names given in the docstring: `model({"input_ids": input_ids, "token_type_ids": token_type_ids})` Note that when creating models and layers with [subclassing](https://keras.io/guides/making_new_layers_and_models_via_subclassing/) then you don't need to worry about any of this, as you can just pass inputs like you would to any other Python function! Parameters: config ([`OpenAIGPTConfig`]): 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 (`Numpy array` or `tf.Tensor` of shape `(batch_size, sequence_length)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.__call__`] and [`PreTrainedTokenizer.encode`] for details. [What are input IDs?](../glossary#input-ids) attention_mask (`tf.Tensor` or `Numpy array` of shape `(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**. [What are attention masks?](../glossary#attention-mask) token_type_ids (`tf.Tensor` or `Numpy array` of shape `(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. [What are token type IDs?](../glossary#token-type-ids) position_ids (`tf.Tensor` or `Numpy array` of shape `(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]`. [What are position IDs?](../glossary#position-ids) head_mask (`tf.Tensor` or `Numpy array` 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 (`tf.Tensor` or `Numpy array` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert `input_ids` indices into associated vectors than the model's internal embedding lookup matrix. output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the config will be used instead. 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. This argument can be used only in eager mode, in graph mode the value in the config will be used instead. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. This argument can be used in eager mode, in graph mode the value will always be set to True. training (`bool`, *optional*, defaults to `False`): Whether or not to use the model in training mode (some modules like dropout modules have different behaviors between training and evaluation). zdThe bare OpenAI GPT transformer model outputting raw hidden-states without any specific head on top.ceZdZfdZeeeeee e  d ddZ ddZ xZ S)TFOpenAIGPTModelcPt||g|i|t|d|_y)Nrr)r,r-rrr;r=rr>r?s r@r-zTFOpenAIGPTModel.__init__s) 3&3F3/]KrA checkpoint output_typerc <|j||||||||| |  } | S)N rrarrrbrr1rrrW)r) r;rrarrrbrr1rrrWrgs r@r|zTFOpenAIGPTModel.call!s>(""))%'/!5##  rAc|jryd|_t|ddNtj|jj 5|jj ddddyy#1swYyxYwNTrr~rrGrrr)rrs r@rzTFOpenAIGPTModel.buildCm ::  4 - 9t//445 -  &&t, - - : - - A11A:rrrC)rrrr-rrOPENAI_GPT_INPUTS_DOCSTRINGr_CHECKPOINT_FOR_DOCr _CONFIG_FOR_DOCr|rrrs@r@rrs L*+FG&%$.28<8<6:377;,0/3&*#(*66  4  1 5*-$! ) H6-rArz OpenAI GPT Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings). ceZdZfdZdZdZeeee e e e  d d dZ dZd dZxZS) TFOpenAIGPTLMHeadModelc^t||g|i|t|d|_d|_y)NrrF)r,r-rrsupports_xla_generationrs r@r-zTFOpenAIGPTLMHeadModel.__init__Ts1 3&3F3/]K',$rAc"|jSrC)rrs r@get_output_embeddingsz,TFOpenAIGPTLMHeadModel.get_output_embeddingsZs((**rAc&|j|yrC)rrs r@set_output_embeddingsz,TFOpenAIGPTLMHeadModel.set_output_embeddings]s !!%(rArc D|j||||||||| |  } | d} |jj| d}d}| (|ddddf}| ddddf} |j| |}| s|f| ddz}||f|zS|St||| j| j S) labels (`tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the cross entropy classification loss. Indices should be in `[0, ..., config.vocab_size - 1]`. rrlinearrNrQr!lossrrr)rrhf_compute_lossr rr)r;rrarrrbrr1rrlabelsrWtransformer_outputsrrrshifted_logitsoutputs r@r|zTFOpenAIGPTLMHeadModel.call`s6#..))%'/!5#/  ,A. !!..}8.L  #AssF^NAqrE]F''?DY!4QR!88F)-)9TGf$ Ev E-;;*55   rAc  d|iS)Nrr+)r;rr>s r@prepare_inputs_for_generationz4TFOpenAIGPTLMHeadModel.prepare_inputs_for_generations V$$rAc|jryd|_t|ddNtj|jj 5|jj ddddyy#1swYyxYwrrrs r@rzTFOpenAIGPTLMHeadModel.buildrr NNNNNNNNNNF)rrrarrrrrrbrrrr1rrrrrrrrWrrzUnion[Tuple, TFCausalLMOutput]rC)rrrr-rrrrrrrr r r|rrrrs@r@r r Ls- +)*+FG&$$.28<8<6:377;,0/3&*04#(4 *4 64 6 4 4 4 1 4 54 *4 -4 $4 .4 !4  (4 H4 l%-rAr a} OpenAI GPT Model transformer with a language modeling and a multiple-choice classification head on top e.g. for RocStories/SWAG tasks. The two heads are two linear layers. The language modeling head has its weights tied to the input embeddings, the classification head takes as input the input of a specified classification token index in the input sequence). ceZdZfdZeeeeee  d ddZ e dZ ddZ xZS) TFOpenAIGPTDoubleHeadsModelct||g|i|d|_t|d|_t ||j d|_y)Nr!rrmultiple_choice_headr')r,r- num_labelsrrrr(r#rs r@r-z$TFOpenAIGPTDoubleHeadsModel.__init__sL 3&3F3/]K$5 f&>&>E[% !rA)rrc | t|} nt|dd} | d} |tj|d| fnd}|tj|d| fnd}|tj|d| fnd}|tj|d| fnd}|j|||||||| | |  }|d}tj|| t|ddz}| r| r|jdd|fz}nd}|jj |d}|j ||| }tj|d}| s ||f|d dzSt||||j S) a mc_token_ids (`tf.Tensor` or `Numpy array` of shape `(batch_size, num_choices)`, *optional*, default to index of the last token of the input): Index of the classification token in each input sequence. Selected in the range `[0, input_ids.size(-1) - 1]`. Return: Examples: ```python >>> import tensorflow as tf >>> from transformers import AutoTokenizer, TFOpenAIGPTDoubleHeadsModel >>> tokenizer = AutoTokenizer.from_pretrained("openai-community/openai-gpt") >>> model = TFOpenAIGPTDoubleHeadsModel.from_pretrained("openai-community/openai-gpt") >>> # Add a [CLS] to the vocabulary (we should train it also!) >>> tokenizer.add_special_tokens({"cls_token": "[CLS]"}) >>> model.resize_token_embeddings(len(tokenizer)) # Update the model embeddings with the new vocabulary size >>> print(tokenizer.cls_token_id, len(tokenizer)) # The newly token the last token of the vocabulary >>> choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"] >>> encoding = tokenizer(choices, return_tensors="tf") >>> inputs = {k: tf.expand_dims(v, 0) for k, v in encoding.items()} >>> inputs["mc_token_ids"] = tf.constant( ... [inputs["input_ids"].shape[-1] - 1, inputs["input_ids"].shape[-1] - 1] ... )[ ... None, : ... ] # Batch size 1 >>> outputs = model(inputs) >>> lm_prediction_scores, mc_prediction_scores = outputs[:2] ```NrQ)rrWrrrrVrTr!)rrrr) rrGr\rrrr#squeezerr)r;rrarrrbr mc_token_idsr1rrrW input_shapes seq_lengthflat_input_idsflat_attention_maskflat_token_type_idsflat_position_idsrrr lm_logitsrs r@r|z TFOpenAIGPTDoubleHeadsModel.callsd  %i0L%m4Sb9L!"% DMDYIJ/?@_cN\Nhbjj"j9IJnrN\Nhbjj"j9IJnrJVJbBJJ|b*5EFhl"..        #/  ,A.  =,MAZ[][^A_2_` /!4 A A#2 F-IY Y  $ $$11-h1O --m\T\-] JJyr2 y),?,CC C0+*55   rActjdtjdtjdtjdtjdtjddS)N)NNNrrra)NNr)rrar')rG TensorSpecint32rs r@input_signaturez+TFOpenAIGPTDoubleHeadsModel.input_signaturesN'9288+V mm,>O_`MM,GWX  rAc|jryd|_t|ddMtj|jj 5|jj ddddt|ddNtj|jj 5|jj ddddyy#1swYexYw#1swYyxYw)NTrr#)r~rrGrrr)rr#rs r@rz!TFOpenAIGPTDoubleHeadsModel.build!s ::  4 - 9t//445 -  &&t, - 4/ 6 Bt88==> 6))//5 6 6 C - - 6 6s C %C CC r)rrrarrrrrrbrrrr'rr1rrrrrrWrrz/Union[Tuple, TFOpenAIGPTDoubleHeadsModelOutput]rC)rrrr-rrrr rr r|propertyr2rrrs@r@r!r!s *+FG+L[jk.28<8<6:377;6:,0/3&*#(Y *Y 6Y 6 Y 4 Y 1 Y 5Y 4Y *Y -Y $Y !Y  9Y lHY v   6rAr!a The OpenAI GPT Model transformer with a sequence classification head on top (linear layer). [`TFOpenAIGPTForSequenceClassification`] uses the last token in order to do the classification, as other causal models (e.g. GPT-2) do. Since it does classification on the last token, it requires to know the position of the last token. If a `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in each row of the batch). ceZdZfdZeeeeee e  d ddZ ddZ xZ S)$TFOpenAIGPTForSequenceClassificationc t||g|i||j|_tjj |jt |jdd|_t|d|_ ||_ y)NscoreF)kernel_initializerr)use_biasrr) r,r-r$rr2Denserr(r8rrr=rs r@r-z-TFOpenAIGPTForSequenceClassification.__init__=sw 3&3F3 ++\\''   .v/G/GH (  0]K rArc 4|j||||||||| |  } | d} |j| }d}|jjd}n|t j t j tjj||jj|jddz }t j|dk\||jddz }t j||dd}n.d}tj|jj dd}| |t#|dd \}}nt#|dd \}}|jj |dk(sJd t j$|s |d||f}|j't j(| ddgt j(|d|j*g}||n|}| s|f| ddz}||f|zS|St-||| j.| j0 S) rrrNrQrTr!) batch_dimsrUz will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`rkz=Cannot handle batch sizes > 1 if no padding token is defined.r)rr8r= pad_token_idrGargmaxrYrZequalrSwhererrlogger warning_oncer?rr is_tensorrr\r$r rr)r;rrarrrbrr1rrrrWrrr in_logitssequence_lengthsr batch_sizesequence_length pooled_logitsrs r@r|z)TFOpenAIGPTForSequenceClassification.callIsS4#..))%'/!5#/  ,A. M* ;; # # +! $IIbggbggmmIt{{?W?W&XZcZiZijqst!$&88,<,ACSU^UdUdegUhklUl#m IIf.>1STU #% ##~~../0^^  $.8.CBQ.G+ O.8.G.K+ O ((4 a ON OG<< 01"1Z<1A#AB '' 6B7(CRZZPY\^`d`o`o[pEqrD%.%:  #%(;AB(??F)-)9TGf$ Ev E) -;;*55   rAc|jryd|_t|dddtj|jj 5|jj dd|jjgdddt|ddNtj|jj 5|jj ddddyy#1swYexYw#1swYyxYw)NTr8r) r~rrGrr8r)rr=rrrs r@rz*TFOpenAIGPTForSequenceClassification.builds ::  4$ ' 3tzz/ C   $dkk.@.@!AB C 4 - 9t//445 -  &&t, - - : C C - -s 3C"<C."C+.C7r)rrrarrrrrrbrrrr1rrrrrrrrWrrz(Union[Tuple, TFSequenceClassifierOutput]rC)rrrr-rrrrrr r r|rrrs@r@r6r6-s *+FG&.$.28<8<6:377;,0/3&*04#(N *N 6N 6 N 4 N 1 N 5N *N -N $N .N !N  2N HN ` -rAr6)?r __future__r dataclassesrtypingrrrnumpynp tensorflowrGactivations_tfr modeling_tf_outputsr r r modeling_tf_utilsr rrrrrrrrrrtf_utilsrrrutilsrrrrrr configuration_openair" get_loggerrrBrr r2Layerr$rrrrrOPENAI_GPT_START_DOCSTRINGrrr r!r6r+rAr@rZs "!))/bb    SR2   H %3#g>%,,$$g>T9ELL  9:$7ell  $7Nb 5<<--b b J&!2& / / /8(T5pj--1-- --` S-79US-S-l x6"<x6x6v m-+EGcm-m-rA