ó
    ˆŽ¡j¢|  ã                   óÀ  • S r SSKrSSKJr  SSKJr  SSKJr  SSKrSSKJ	r	  SSK
JrJrJr  SS	KJr  SS
KJrJrJr  SSKJr  SSKJrJrJr  SSKJr  SSKJr  SSKJ r J!r!J"r"  SSK#J$r$  \"RJ                  " \&5      r'\	RP                  " 5       \\\S.r) " S S\	RT                  5      r+ " S S\	RT                  5      r, " S S\	RT                  5      r- " S S\	RT                  5      r.\! " S S\5      5       r/\!" SS9\ " S S \ 5      5       5       r0\! " S! S"\/5      5       r1\!" S#S9 " S$ S%\/\5      5       r2\!" S&S9 " S' S(\/5      5       r3\!" S)S9 " S* S+\/5      5       r4/ S,Qr5g)-zPyTorch OpenAI GPT model.é    N)ÚCallable)Ú	dataclass)ÚAny)Únn)ÚBCEWithLogitsLossÚCrossEntropyLossÚMSELossé   )Úinitialization)Úgelu_newÚget_activationÚsilu)ÚGenerationMixin)ÚBaseModelOutputÚCausalLMOutputÚSequenceClassifierOutput)ÚPreTrainedModel)ÚConv1D)ÚModelOutputÚauto_docstringÚloggingé   )ÚOpenAIGPTConfig)Úrelur   ÚgeluÚswishc                   óP   ^ • \ rS rSrSU 4S jjrS	S jrS rSS jrS	S jrSr	U =r
$ )
Ú	Attentioné.   c                 óH  >• [         TU ]  5         X l        UnXSR                  -  S:w  a  [	        SU SUR                   35      e[
        R                  " [        R                  " [        R                  " X"5      5      R                  SSX"5      SS9U l        UR                  U l        XPl        X@l        [        US-  U5      U l        [        XQ5      U l        [
        R"                  " UR$                  5      U l        [
        R"                  " UR(                  5      U l        g )Nr   zAttention n_state shape: z$ must be divisible by config.n_head r   F©Ú
persistentr
   )ÚsuperÚ__init__Ún_positionsÚn_headÚ
ValueErrorr   ÚBufferÚtorchÚtrilÚonesÚviewÚbiasÚ
split_sizeÚscaler   Úc_attnÚc_projÚDropoutÚ
attn_pdropÚattn_dropoutÚresid_pdropÚresid_dropout)ÚselfÚnxr%   Úconfigr/   Ún_stateÚ	__class__s         €Ú¸C:\Users\johnn\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages\transformers/models/openai/modeling_openai.pyr$   ÚAttention.__init__/   sä   ø€ Ü‰ÑÔØ&ÔØˆØ—]‘]Ñ" aÓ'ÜÐ8¸¸	ÐAeÐfl×fsÑfsÐetÐuÓvÐvÜ—I’IÜ�JŠJ”u—z’z +Ó;Ó<×AÑAÀ!ÀQÈÓaÐnsñ
ˆŒ	ð —m‘mˆŒØ!ŒØŒ
ä˜W q™[¨"Ó-ˆŒÜ˜WÓ)ˆŒÜŸJšJ v×'8Ñ'8Ó9ˆÔÜŸZšZ¨×(:Ñ(:Ó;ˆÕó    c                 óú  • [         R                  " X5      nU R                  (       a(  U[        R                  " UR                  S5      5      -  nU R                  S S 2S S 2S UR                  S5      2S UR                  S5      24   nXg-  SSU-
  -  -   nUb  Xd-   n[        R                  R                  USS9nU R                  U5      n[         R                  " Xc5      /nU(       a  UR                  U5        U$ )Néÿÿÿÿéþÿÿÿg     ˆÃÀr   ©Údim)r)   Úmatmulr/   ÚmathÚsqrtÚsizer-   r   Ú
functionalÚsoftmaxr4   Úappend)	r7   ÚqÚkÚvÚattention_maskÚoutput_attentionsÚwÚbÚoutputss	            r<   Ú_attnÚAttention._attnA   sÑ   € Ü�LŠL˜ÓˆØ�:�:Ø”D—I’I˜aŸf™f R›jÓ)Ñ)ˆAà�I‰I’aš˜L˜aŸf™f R›j˜L¨,¨A¯F©F°2«J¨,Ð6Ñ7ˆØ‰E�D˜A ™E‘NÑ"ˆàÑ%àÑ"ˆAä�M‰M×!Ñ! !¨Ð!Ð,ˆØ×Ñ˜aÓ ˆä—<’< Ó%Ð&ˆÞØ�N‰N˜1ÔØˆr>   c                 óÔ   • UR                  SSSS5      R                  5       nUR                  5       S S UR                  S5      UR                  S5      -  4-   nUR                  " U6 $ )Nr   é   r   r
   rA   r@   )ÚpermuteÚ
contiguousrG   r,   )r7   ÚxÚnew_x_shapes      r<   Úmerge_headsÚAttention.merge_headsU   s\   € Ø�I‰I�a˜˜A˜qÓ!×,Ñ,Ó.ˆØ—f‘f“h˜s �m q§v¡v¨b£z°A·F±F¸2³JÑ'>Ð&@Ñ@ˆØ�vŠv�{Ð#Ð#r>   c                 óú   • UR                  5       S S U R                  UR                  S5      U R                  -  4-   nUR                  " U6 nU(       a  UR                  SSSS5      $ UR                  SSSS5      $ )Nr@   r   rV   r
   r   )rG   r&   r,   rW   )r7   rY   rL   rZ   s       r<   Úsplit_headsÚAttention.split_headsZ   sm   € Ø—f‘f“h˜s �m t§{¡{°A·F±F¸2³JÀ$Ç+Á+Ñ4MÐ&NÑNˆØ�FŠF�KÐ ˆÞØ—9‘9˜Q  1 aÓ(Ð(à—9‘9˜Q  1 aÓ(Ð(r>   c                 ón  • U R                  U5      nUR                  U R                  SS9u  pEnU R                  U5      nU R                  USS9nU R                  U5      nU R	                  XEXbU5      nUS   nU R                  U5      nU R                  U5      nU R                  U5      nU/USS  -   n	U	$ )NrV   rB   T)rL   r   r   )r0   Úsplitr.   r^   rS   r[   r1   r6   )
r7   rY   rN   rO   ÚqueryÚkeyÚvalueÚattn_outputsÚarR   s
             r<   ÚforwardÚAttention.forwardb   sÀ   € Ø�K‰K˜‹NˆØŸG™G D§O¡O¸˜GÐ;Ñˆ�EØ× Ñ  Ó'ˆØ×Ñ˜s dÐÐ+ˆØ× Ñ  Ó'ˆà—z‘z %¨eÐEVÓWˆØ˜‰Oˆà×Ñ˜QÓˆØ�K‰K˜‹NˆØ×Ñ˜qÓ!ˆà�#˜ Q RÐ(Ñ(ˆØˆr>   )	r4   r-   r0   r1   r&   r%   r6   r/   r.   ©F©NF)Ú__name__Ú
__module__Ú__qualname__Ú__firstlineno__r$   rS   r[   r^   rg   Ú__static_attributes__Ú__classcell__©r;   s   @r<   r   r   .   s!   ø† ÷<ô$ò($ô
)÷ò r>   r   c                   ó.   ^ • \ rS rSrU 4S jrS rSrU =r$ )ÚMLPét   c                 óô   >• [         TU ]  5         UR                  n[        X5      U l        [        X15      U l        [        UR                     U l        [        R                  " UR                  5      U l        g ©N)r#   r$   Ún_embdr   Úc_fcr1   ÚACT_FNSÚafnÚactr   r2   r5   Údropout)r7   r:   r9   r8   r;   s       €r<   r$   ÚMLP.__init__u   sU   ø€ Ü‰ÑÔØ�]‰]ˆÜ˜7Ó'ˆŒ	Ü˜RÓ)ˆŒÜ˜6Ÿ:™:Ñ&ˆŒÜ—z’z &×"4Ñ"4Ó5ˆ�r>   c                 ó†   • U R                  U R                  U5      5      nU R                  U5      nU R                  U5      $ rv   ©r{   rx   r1   r|   )r7   rY   ÚhÚh2s       r<   rg   ÚMLP.forward}   s4   € Ø�H‰H�T—Y‘Y˜q“\Ó"ˆØ�[‰[˜‹^ˆØ�|‰|˜BÓÐr>   r   ©rk   rl   rm   rn   r$   rg   ro   rp   rq   s   @r<   rs   rs   t   s   ø† õ6÷ ð  r>   rs   c                   ó6   ^ • \ rS rSrSU 4S jjrSS jrSrU =r$ )ÚBlockéƒ   c                 ó  >• [         TU ]  5         UR                  n[        XAX#5      U l        [
        R                  " XBR                  S9U l        [        SU-  U5      U l
        [
        R                  " XBR                  S9U l        g )N)Úepsé   )r#   r$   rw   r   Úattnr   Ú	LayerNormÚlayer_norm_epsilonÚln_1rs   ÚmlpÚln_2)r7   r%   r9   r/   r8   r;   s        €r<   r$   ÚBlock.__init__„   sc   ø€ Ü‰ÑÔØ�]‰]ˆÜ˜b¨vÓ=ˆŒ	Ü—L’L ×)BÑ)BÑCˆŒ	Ü�q˜2‘v˜vÓ&ˆŒÜ—L’L ×)BÑ)BÑCˆ�	r>   c                 ó²   • U R                  UUUS9nUS   nU R                  X-   5      nU R                  U5      nU R                  Xg-   5      nU/USS  -   n	U	$ )N)rN   rO   r   r   )rŠ   r�   rŽ   r�   )
r7   rY   rN   rO   re   rf   ÚnÚmr€   rR   s
             r<   rg   ÚBlock.forwardŒ   sn   € Ø—y‘yØØ)Ø/ð !ð 
ˆð
 ˜‰Oˆà�I‰I�a‘eÓˆØ�H‰H�Q‹KˆØ�I‰I�a‘eÓˆà�#˜ Q RÐ(Ñ(ˆØˆr>   )rŠ   r�   r�   rŽ   ri   rj   rƒ   rq   s   @r<   r…   r…   ƒ   s   ø† ÷D÷ò r>   r…   c                   ó’   ^ • \ rS rSrSrS\4U 4S jjr SS\R                  S\R                  S-  S\R                  4S	 jjr
S
rU =r$ )ÚOpenAIGPTSequenceSummaryé�   a…  
Compute a single vector summary of a sequence hidden states.

Args:
    config ([`OpenAIGPTConfig`]):
        The config used by the model. Relevant arguments in the config class of the model are (refer to the actual
        config class of your model for the default values it uses):

        - **summary_type** (`str`) -- The method to use to make this summary. Accepted values are:

            - `"last"` -- Take the last token hidden state (like XLNet)
            - `"first"` -- Take the first token hidden state (like Bert)
            - `"mean"` -- Take the mean of all tokens hidden states
            - `"cls_index"` -- Supply a Tensor of classification token position (GPT/GPT-2)
            - `"attn"` -- Not implemented now, use multi-head attention

        - **summary_use_proj** (`bool`) -- Add a projection after the vector extraction.
        - **summary_proj_to_labels** (`bool`) -- If `True`, the projection outputs to `config.num_labels` classes
          (otherwise to `config.hidden_size`).
        - **summary_activation** (`Optional[str]`) -- Set to `"tanh"` to add a tanh activation to the output,
          another string or `None` will add no activation.
        - **summary_first_dropout** (`float`) -- Optional dropout probability before the projection and activation.
        - **summary_last_dropout** (`float`)-- Optional dropout probability after the projection and activation.
r9   c                 óÂ  >• [         TU ]  5         [        USS5      U l        U R                  S:X  a  [        e[
        R                  " 5       U l        [        US5      (       a‚  UR                  (       aq  [        US5      (       a.  UR                  (       a  UR                  S:”  a  UR                  nOUR                  n[
        R                  " UR                  U5      U l        [        USS 5      nU(       a  [        U5      O[
        R                  " 5       U l        [
        R                  " 5       U l        [        US5      (       a5  UR"                  S:”  a%  [
        R$                  " UR"                  5      U l        [
        R                  " 5       U l        [        US	5      (       a7  UR(                  S:”  a&  [
        R$                  " UR(                  5      U l        g g g )
NÚsummary_typeÚlastrŠ   Úsummary_use_projÚsummary_proj_to_labelsr   Úsummary_activationÚsummary_first_dropoutÚsummary_last_dropout)r#   r$   Úgetattrr™   ÚNotImplementedErrorr   ÚIdentityÚsummaryÚhasattrr›   rœ   Ú
num_labelsÚhidden_sizeÚLinearr   Ú
activationÚfirst_dropoutrž   r2   Úlast_dropoutrŸ   )r7   r9   Únum_classesÚactivation_stringr;   s       €r<   r$   Ú!OpenAIGPTSequenceSummary.__init__·   sa  ø€ Ü‰ÑÔä# F¨N¸FÓCˆÔØ×Ñ Ó&ô &Ð%ä—{’{“}ˆŒÜ�6Ð-×.Ñ.°6×3J×3JÜ�vÐ7×8Ñ8¸V×=Z×=ZÐ_e×_pÑ_pÐstÓ_tØ$×/Ñ/‘à$×0Ñ0�ÜŸ9š9 V×%7Ñ%7¸ÓEˆDŒLä# FÐ,@À$ÓGÐÞIZ¤NÐ3DÔ$EÔ`b×`kÒ`kÓ`mˆŒäŸ[š[›]ˆÔÜ�6Ð2×3Ñ3¸×8TÑ8TÐWXÓ8XÜ!#§¢¨F×,HÑ,HÓ!IˆDÔäŸKšK›MˆÔÜ�6Ð1×2Ñ2°v×7RÑ7RÐUVÓ7VÜ "§
¢
¨6×+FÑ+FÓ GˆDÕð 8WÐ2r>   NÚhidden_statesÚ	cls_indexÚreturnc                 óþ  • U R                   S:X  a  USS2S4   nGOU R                   S:X  a  USS2S4   nGOU R                   S:X  a  UR                  SS9nOâU R                   S	:X  a¼  Uc?  [        R                  " US
SS2SS24   UR                  S   S-
  [        R
                  S9nOXUR                  S5      R                  S5      nUR                  SUR                  5       S-
  -  UR                  S5      4-   5      nUR                  SU5      R                  S5      nOU R                   S:X  a  [        eU R                  W5      nU R                  U5      nU R                  U5      nU R!                  U5      nU$ )a#  
Compute a single vector summary of a sequence hidden states.

Args:
    hidden_states (`torch.FloatTensor` of shape `[batch_size, seq_len, hidden_size]`):
        The hidden states of the last layer.
    cls_index (`torch.LongTensor` of shape `[batch_size]` or `[batch_size, ...]` where ... are optional leading dimensions of `hidden_states`, *optional*):
        Used if `summary_type == "cls_index"` and takes the last token of the sequence as classification token.

Returns:
    `torch.FloatTensor`: The summary of the sequence hidden states.
rš   Nr@   Úfirstr   Úmeanr   rB   r¯   .rA   ©Údtype)r@   rŠ   )r™   r³   r)   Ú	full_likeÚshapeÚlongÚ	unsqueezeÚexpandrC   rG   ÚgatherÚsqueezer¡   r©   r£   r¨   rª   )r7   r®   r¯   Úoutputs       r<   rg   Ú OpenAIGPTSequenceSummary.forwardÔ   sn  € ð ×Ñ Ó&Ø"¢1 b 5Ñ)ŠFØ×Ñ 'Ó)Ø"¢1 a 4Ñ(ŠFØ×Ñ &Ó(Ø"×'Ñ'¨AÐ'Ð.‰FØ×Ñ +Ó-ØÑ Ü!ŸOšOØ! # r¨ rª1 *Ñ-Ø!×'Ñ'¨Ñ+¨aÑ/ÜŸ*™*ñ‘	ð &×/Ñ/°Ó3×=Ñ=¸bÓA�	Ø%×,Ñ,¨U°i·m±m³oÈÑ6IÑ-JÈm×N`ÑN`ÐacÓNdÐMfÑ-fÓg�	à"×)Ñ)¨"¨iÓ8×@Ñ@ÀÓD‰FØ×Ñ &Ó(Ü%Ð%à×#Ñ# FÓ+ˆØ—‘˜fÓ%ˆØ—‘ Ó(ˆØ×"Ñ" 6Ó*ˆàˆr>   )r¨   r©   rª   r£   r™   rv   )rk   rl   rm   rn   Ú__doc__r   r$   r)   ÚFloatTensorÚ
LongTensorrg   ro   rp   rq   s   @r<   r–   r–   �   sV   ø† ñð2H˜÷ Hð< VZñ)Ø"×.Ñ.ð)Ø;@×;KÑ;KÈdÑ;Rð)à	×	Ñ	÷)ó )r>   r–   c                   ó8   ^ • \ rS rSr% \\S'   SrU 4S jrSrU =r	$ )ÚOpenAIGPTPreTrainedModelé   r9   Útransformerc           	      óÚ  >• [         TU ]  U5        [        U[        5      (       ag  UR                  n[
        R                  " UR                  [        R                  " [        R                  " X"5      5      R                  SSX"5      5        g [        U[        5      (       aJ  [
        R                  " UR                  [        R                  " UR                  R                  5      5        g g )Nr   )r#   Ú_init_weightsÚ
isinstancer   r%   ÚinitÚcopy_r-   r)   r*   r+   r,   ÚOpenAIGPTModelÚposition_idsÚaranger9   )r7   Úmoduler%   r;   s      €r<   rÇ   Ú&OpenAIGPTPreTrainedModel._init_weights  sœ   ø€ Ü‰Ñ˜fÔ%Ü�fœi×(Ñ(Ø ×,Ñ,ˆKÜ�JŠJØ—‘œUŸZšZ¬¯
ª
°;Ó(LÓM×RÑRÐSTÐVWÐYdÓrõô ˜¤×/Ñ/Ü�JŠJ�v×*Ñ*¬E¯LªL¸¿¹×9RÑ9RÓ,SÕTð 0r>   © )
rk   rl   rm   rn   r   Ú__annotations__Úbase_model_prefixrÇ   ro   rp   rq   s   @r<   rÃ   rÃ      s   ø‡ àÓØ%Ð÷Uó Ur>   rÃ   z^
    Base class for outputs of models predicting if two sentences are consecutive or not.
    )Úcustom_introc                   ó  • \ rS rSr% SrSr\R                  S-  \S'   Sr	\R                  S-  \S'   Sr
\R                  S-  \S'   Sr\R                  S-  \S'   Sr\\R                     S-  \S'   Sr\\R                     S-  \S	'   S
rg)ÚOpenAIGPTDoubleHeadsModelOutputi  a‚  
loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
    Language modeling loss.
mc_loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `mc_labels` is provided):
    Multiple choice classification loss.
logits (`torch.FloatTensor` of shape `(batch_size, num_choices, sequence_length, config.vocab_size)`):
    Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
mc_logits (`torch.FloatTensor` of shape `(batch_size, num_choices)`):
    Prediction scores of the multiple choice classification head (scores for each choice before SoftMax).
NÚlossÚmc_lossÚlogitsÚ	mc_logitsr®   Ú
attentionsrÐ   )rk   rl   rm   rn   r¿   rÖ   r)   rÀ   rÑ   r×   rØ   rÙ   r®   ÚtuplerÚ   ro   rÐ   r>   r<   rÕ   rÕ     s‘   ‡ ñ	ð &*€Dˆ%×
Ñ
˜dÑ
"Ó)Ø(,€GˆU×Ñ Ñ%Ó,Ø'+€FˆE×Ñ Ñ$Ó+Ø*.€Iˆu× Ñ  4Ñ'Ó.Ø59€M�5˜×*Ñ*Ñ+¨dÑ2Ó9Ø26€J��e×'Ñ'Ñ(¨4Ñ/Ö6r>   rÕ   c                   ó4  ^ • \ rS rSrU 4S jrS rS r\        SS\R                  S-  S\R                  S-  S\R                  S-  S	\R                  S-  S
\R                  S-  S\S-  S\S-  S\S-  S\\R                     \-  4S jj5       rSrU =r$ )rË   i*  c                 ól  >• [         TU ]  U5        [        R                  " UR                  UR
                  5      U l        [        R                  " UR                  UR
                  5      U l        [        R                  " UR                  5      U l        [        R                  " [        UR                  5       Vs/ s H  n[        UR                  USS9PM     sn5      U l        [        R"                  " [$        R&                  " UR                  5      SS9U l        U R+                  5         g s  snf )NT)r/   Fr!   )r#   r$   r   Ú	EmbeddingÚ
vocab_sizerw   Útokens_embedr%   Úpositions_embedr2   Ú
embd_pdropÚdropÚ
ModuleListÚrangeÚn_layerr…   r€   r(   r)   rÍ   rÌ   Ú	post_init)r7   r9   Ú_r;   s      €r<   r$   ÚOpenAIGPTModel.__init__,  sÑ   ø€ Ü‰Ñ˜Ô äŸLšL¨×):Ñ):¸F¿M¹MÓJˆÔÜ!Ÿ|š|¨F×,>Ñ,>ÀÇÁÓNˆÔÜ—J’J˜v×0Ñ0Ó1ˆŒ	Ü—’ÔW\Ð]c×]kÑ]kÔWlÓmÒWlÐRS¤ f×&8Ñ&8¸&ÈÔ MÑWlÑmÓnˆŒäŸIšI¤e§l¢l°6×3EÑ3EÓ&FÐSXÑYˆÔà�‰Õùò	  ns   Â=D1c                 ó   • U R                   $ rv   ©rà   )r7   s    r<   Úget_input_embeddingsÚ#OpenAIGPTModel.get_input_embeddings8  s   € Ø× Ñ Ð r>   c                 ó   • Xl         g rv   rë   )r7   Únew_embeddingss     r<   Úset_input_embeddingsÚ#OpenAIGPTModel.set_input_embeddings;  s   € Ø*Õr>   NÚ	input_idsrN   Útoken_type_idsrÌ   Úinputs_embedsrO   Úoutput_hidden_statesÚreturn_dictr°   c	                 ó  • Ub  UOU R                   R                  nUb  UOU R                   R                  nUb  UOU R                   R                  nUb  Ub  [	        S5      eUb7  U R                  X5        UR                  5       n
UR                  SU
S   5      nO"Ub  UR                  5       S S n
O[	        S5      eUc  U R                  S S U
S   24   nUb€  UR                  S5      R                  S5      nUR                  [        U R                  5       5      R                  S9nSU-
  [        R                  " U R                  5      R                   -  nUc  U R#                  U5      nU R%                  U5      nUb3  UR                  SUR                  S5      5      nU R#                  U5      nOSnX[-   U-   nU R'                  U5      nX­R                  S5      4-   nU(       a  S	OS nU(       a  S	OS n[)        U R*                  5       H1  u  nnU(       a  UU4-   nU" XÒUS
9nUS   nU(       d  M(  UUS   4-   nM3     UR                  " U6 nU(       a  UU4-   nU(       d  [-        S UUU4 5       5      $ [/        UUUS9$ )NzDYou cannot specify both input_ids and inputs_embeds at the same timer@   z5You have to specify either input_ids or inputs_embedsr   rV   r´   g      ð?r   rÐ   )rO   c              3   ó.   #   • U  H  oc  M  Uv •  M     g 7frv   rÐ   )Ú.0rM   s     r<   Ú	<genexpr>Ú)OpenAIGPTModel.forward.<locals>.<genexpr>�  s   é € ÐhÒ$V˜qŸ™Ò$Vùs   ‚Œ	)Úlast_hidden_stater®   rÚ   )r9   rO   rõ   rö   r'   Ú%warn_if_padding_and_no_attention_maskrG   r,   rÌ   r¹   ÚtoÚnextÚ
parametersrµ   r)   ÚfinfoÚminrà   rá   rã   Ú	enumerater€   rÛ   r   )r7   rò   rN   ró   rÌ   rô   rO   rõ   rö   ÚkwargsÚinput_shapeÚposition_embedsÚtoken_type_embedsr®   Úoutput_shapeÚall_attentionsÚall_hidden_statesÚiÚblockrR   s                       r<   rg   ÚOpenAIGPTModel.forward>  s¥  € ð 2CÑ1NÑ-ÐTX×T_ÑT_×TqÑTqÐà$8Ñ$DÑ È$Ï+É+×JjÑJjð 	ð &1Ñ%<‘kÀ$Ç+Á+×BYÑBYˆàÑ  ]Ñ%>ÜÐcÓdÐdØÑ"Ø×6Ñ6°yÔQØ#Ÿ.™.Ó*ˆKØ!Ÿ™ r¨;°r©?Ó;‰IØÑ&Ø'×,Ñ,Ó.¨s°Ð3‰KäÐTÓUÐUàÑà×,Ñ,¨TÐ3D°[À±_Ð3DÐ-DÑEˆLð Ñ%ð ,×5Ñ5°aÓ8×BÑBÀ1ÓEˆNð ,×.Ñ.´T¸$¿/¹/Ó:KÓ5L×5RÑ5RÐ.ÐSˆNØ! NÑ2´e·k²kÀ$Ç*Á*Ó6M×6QÑ6QÑQˆNàÑ Ø ×-Ñ-¨iÓ8ˆMØ×.Ñ.¨|Ó<ˆØÑ%Ø+×0Ñ0°°^×5HÑ5HÈÓ5LÓMˆNØ $× 1Ñ 1°.Ó AÑà !ÐØ%Ñ7Ð:KÑKˆØŸ	™	 -Ó0ˆà"×&8Ñ&8¸Ó&<Ð%>Ñ>ˆæ0™°dˆÞ"6™B¸DÐÜ! $§&¡&Ö)‰HˆAˆuÞ#Ø$5¸Ð8HÑ$HÐ!á˜MÐM^Ñ_ˆGØ# A™JˆMß Ð Ø!/°7¸1±:°-Ñ!?’ñ *ð &×*Ò*¨LÐ9ˆæØ 1°]Ð4DÑ DÐæÜÑh ]Ð4EÀ~Ñ$VÓhÓhÐhäØ+Ø+Ø%ñ
ð 	
r>   )rã   r€   rÌ   rá   rà   )NNNNNNNN)rk   rl   rm   rn   r$   rì   rð   r   r)   rÁ   rÀ   ÚboolrÛ   ÚTensorr   rg   ro   rp   rq   s   @r<   rË   rË   *  sô   ø† õ
ò!ò+ð ð .2Ø37Ø26Ø04Ø26Ø)-Ø,0Ø#'ñV
à×#Ñ# dÑ*ðV
ð ×)Ñ)¨DÑ0ðV
ð ×(Ñ(¨4Ñ/ð	V
ð
 ×&Ñ&¨Ñ-ðV
ð ×(Ñ(¨4Ñ/ðV
ð   $™;ðV
ð # T™kðV
ð ˜D‘[ðV
ð 
ˆu�|‰|Ñ	˜Ñ	.ôV
ó öV
r>   rË   z‰
    OpenAI GPT Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    c                   ó   ^ • \ rS rSrSS0rU 4S jr\          SS\R                  S-  S\R                  S-  S\R                  S-  S	\R                  S-  S
\R                  S-  S\R                  S-  S\
S-  S\
S-  S\
S-  S\\R                  -  S\\R                     \-  4S jj5       rS\R                  S\\\4   4S jrSrU =r$ )ÚOpenAIGPTLMHeadModeli˜  úlm_head.weightútransformer.tokens_embed.weightc                 óÂ   >• [         TU ]  U5        [        U5      U l        [        R
                  " UR                  UR                  SS9U l        U R                  5         g ©NF©r-   )
r#   r$   rË   rÅ   r   r§   rw   rß   Úlm_headrç   ©r7   r9   r;   s     €r<   r$   ÚOpenAIGPTLMHeadModel.__init__¡  sG   ø€ Ü‰Ñ˜Ô Ü)¨&Ó1ˆÔÜ—y’y §¡°×0AÑ0AÈÑNˆŒð 	�‰Õr>   Nrò   rN   ró   rÌ   rô   ÚlabelsrO   rõ   rö   Úlogits_to_keepr°   c                 ó¾  • U	b  U	OU R                   R                  n	U R                  UUUUUUUU	S9nUS   n[        U
[        5      (       a  [        U
* S5      OU
nU R                  USS2USS24   5      nSnUb)  U R                  " SXöU R                   R                  S.UD6nU	(       d  U4USS -   nUb  U4U-   $ U$ [        UUUR                  UR                  S9$ )a‹  
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
    Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
    `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
    are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
N©rN   ró   rÌ   rô   rO   rõ   rö   r   )rØ   r  rß   r   ©rÖ   rØ   r®   rÚ   rÐ   )r9   rö   rÅ   rÈ   ÚintÚslicer  Úloss_functionrß   r   r®   rÚ   )r7   rò   rN   ró   rÌ   rô   r  rO   rõ   rö   r  r  Útransformer_outputsr®   Úslice_indicesrØ   rÖ   r½   s                     r<   rg   ÚOpenAIGPTLMHeadModel.forward©  s  € ð* &1Ñ%<‘kÀ$Ç+Á+×BYÑBYˆà"×.Ñ.ØØ)Ø)Ø%Ø'Ø/Ø!5Ø#ð /ð 	
Ðð ,¨AÑ.ˆä8BÀ>ÔSV×8WÑ8Wœ˜~˜o¨tÔ4Ð]kˆØ—‘˜mªA¨}ºaÐ,?Ñ@ÓAˆàˆØÑØ×%Ò%Ðp¨VÈtÏ{É{×OeÑOeÑpÐioÑpˆDæØ�YÐ!4°Q°RÐ!8Ñ8ˆFØ)-Ñ)9�T�G˜fÑ$ÐE¸vÐEäØØØ-×;Ñ;Ø*×5Ñ5ñ	
ð 	
r>   c                 óV   • SU0nUR                  5        H  u  pEXC;  d  M  XSU'   M     U$ )Nrò   )Úitems)r7   rò   r  Úmodel_inputsrc   rd   s         r<   Úprepare_inputs_for_generationÚ2OpenAIGPTLMHeadModel.prepare_inputs_for_generationß  s6   € à# YÐ/ˆð !Ÿ,™,ž.‰JˆCØÕ&Ø$)˜SÓ!ñ )ð Ðr>   )r  rÅ   )
NNNNNNNNNr   )rk   rl   rm   rn   Ú_tied_weights_keysr$   r   r)   rÁ   rÀ   r  r  r  rÛ   r   rg   ÚdictÚstrr   r(  ro   rp   rq   s   @r<   r  r  ˜  sG  ø† ð +Ð,MÐNÐõð ð .2Ø37Ø26Ø04Ø26Ø*.Ø)-Ø,0Ø#'Ø-.ñ3
à×#Ñ# dÑ*ð3
ð ×)Ñ)¨DÑ0ð3
ð ×(Ñ(¨4Ñ/ð	3
ð
 ×&Ñ&¨Ñ-ð3
ð ×(Ñ(¨4Ñ/ð3
ð × Ñ  4Ñ'ð3
ð   $™;ð3
ð # T™kð3
ð ˜D‘[ð3
ð ˜eŸl™lÑ*ð3
ð 
ˆu�|‰|Ñ	˜~Ñ	-ô3
ó ð3
ðj	°u×7GÑ7Gð 	ÐVZÐ[^Ð`cÐ[cÑVd÷ 	ò 	r>   r  a�  
        OpenAI GPT Model transformer with a language modeling and a multiple-choice classification head on top e.g. for
    RocStories/SWAG tasks. The two heads are two linear layers. The language modeling head has its weights tied to the
    input embeddings, the classification head takes as input the input of a specified classification token index in the
    input sequence).
    c                   ó�  ^ • \ rS rSrSS0rU 4S jr\           SS\R                  S-  S\R                  S-  S\R                  S-  S	\R                  S-  S
\R                  S-  S\R                  S-  S\R                  S-  S\R                  S-  S\
S-  S\
S-  S\
S-  S\\R                     \-  4S jj5       rSrU =r$ )ÚOpenAIGPTDoubleHeadsModelië  r  r  c                 óð   >• [         TU ]  U5        SUl        [        U5      U l        [
        R                  " UR                  UR                  SS9U l	        [        U5      U l        U R                  5         g )Nr   Fr  )r#   r$   r¥   rË   rÅ   r   r§   rw   rß   r  r–   Úmultiple_choice_headrç   r  s     €r<   r$   Ú"OpenAIGPTDoubleHeadsModel.__init__ö  s\   ø€ Ü‰Ñ˜Ô àˆÔÜ)¨&Ó1ˆÔÜ—y’y §¡°×0AÑ0AÈÑNˆŒÜ$<¸VÓ$DˆÔ!ð 	�‰Õr>   Nrò   rN   ró   rÌ   rô   Úmc_token_idsr  Ú	mc_labelsrO   rõ   rö   r°   c                 óÚ  • Ub  UOU R                   R                  nU R                  UUUUUU	U
US9nUS   nU R                  U5      nU R	                  Xæ5      R                  S5      nSu  nnUbA  [        5       nU" UR                  SUR                  S5      5      UR                  S5      5      nUbr  USSS2SS24   R                  5       nUSSS24   R                  5       n[        5       nU" UR                  SUR                  S5      5      UR                  S5      5      nU(       d  UU4USS -   nUb  U4U-   nUb  U4U-   $ U$ [        UUUUUR                  UR                  S9$ )	a  
mc_token_ids (`torch.LongTensor` 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]`.
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
    Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
    `labels = input_ids` Indices are selected in `[-1, 0, ..., config.vocab_size]` All labels set to `-100` are
    ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
mc_labels (`torch.LongTensor` of shape `(batch_size)`, *optional*):
    Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices]`
    where *num_choices* is the size of the second dimension of the input tensors. (see *input_ids* above)

Examples:

```python
>>> from transformers import AutoTokenizer, OpenAIGPTDoubleHeadsModel
>>> import torch

>>> tokenizer = AutoTokenizer.from_pretrained("openai-community/openai-gpt")
>>> model = OpenAIGPTDoubleHeadsModel.from_pretrained("openai-community/openai-gpt")
>>> tokenizer.add_special_tokens(
...     {"cls_token": "[CLS]"}
... )  # Add a [CLS] to the vocabulary (we should train it also!)
>>> model.resize_token_embeddings(len(tokenizer))

>>> choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"]
>>> input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0)  # Batch size 1, 2 choices
>>> mc_token_ids = torch.tensor([input_ids.size(-1) - 1, input_ids.size(-1) - 1]).unsqueeze(0)  # Batch size 1

>>> outputs = model(input_ids, mc_token_ids=mc_token_ids)
>>> lm_logits = outputs.logits
>>> mc_logits = outputs.mc_logits
```Nr  r   r@   )NN.r   )rÖ   r×   rØ   rÙ   r®   rÚ   )r9   rö   rÅ   r  r0  r¼   r   r,   rG   rX   rÕ   r®   rÚ   )r7   rò   rN   ró   rÌ   rô   r2  r  r3  rO   rõ   rö   r  r"  r®   Ú	lm_logitsrÙ   Úlm_lossr×   Úloss_fctÚshift_logitsÚshift_labelsr½   s                          r<   rg   Ú!OpenAIGPTDoubleHeadsModel.forward  s¢  € ðb &1Ñ%<‘kÀ$Ç+Á+×BYÑBYˆà"×.Ñ.ØØ)Ø)Ø%Ø'Ø/Ø!5Ø#ð /ð 	
Ðð ,¨AÑ.ˆà—L‘L Ó/ˆ	Ø×-Ñ-¨mÓJ×RÑRÐSUÓVˆ	à%Ñˆ�ØÑ Ü'Ó)ˆHÙ˜yŸ~™~¨b°)·.±.ÀÓ2DÓEÀyÇ~Á~ÐVXÓGYÓZˆGØÑØ$ S¨#¨2¨#ªq [Ñ1×<Ñ<Ó>ˆLØ! # q¡r '™?×5Ñ5Ó7ˆLÜ'Ó)ˆHÙ˜|×0Ñ0°°\×5FÑ5FÀrÓ5JÓKÈ\×M^ÑM^Ð_aÓMbÓcˆGæØ Ð+Ð.AÀ!À"Ð.EÑEˆFØÑ"Ø!˜ fÑ,�Ø,3Ñ,?�W�J Ñ'ÐKÀVÐKä.ØØØØØ-×;Ñ;Ø*×5Ñ5ñ
ð 	
r>   )r  r0  rÅ   )NNNNNNNNNNN)rk   rl   rm   rn   r*  r$   r   r)   rÁ   rÀ   r  rÛ   r  rÕ   rg   ro   rp   rq   s   @r<   r.  r.  ë  sD  ø† ð <Ð=MÐNÐõ	ð ð .2Ø37Ø26Ø04Ø26Ø04Ø*.Ø-1Ø)-Ø,0Ø#'ñX
à×#Ñ# dÑ*ðX
ð ×)Ñ)¨DÑ0ðX
ð ×(Ñ(¨4Ñ/ð	X
ð
 ×&Ñ&¨Ñ-ðX
ð ×(Ñ(¨4Ñ/ðX
ð ×&Ñ&¨Ñ-ðX
ð × Ñ  4Ñ'ðX
ð ×#Ñ# dÑ*ðX
ð   $™;ðX
ð # T™kðX
ð ˜D‘[ðX
ð 
ˆu�|‰|Ñ	Ð>Ñ	>ôX
ó öX
r>   r.  aâ  
    The Original OpenAI GPT Model transformer with a sequence classification head on top (linear layer).
    [`OpenAIGPTForSequenceClassification`] uses the last token in order to do the classification, as other causal
    models (e.g. GPT-2) do. Since it does classification on the last token, it requires to know the position of the
    last token. If a `pad_token_id` is defined in the configuration, it finds the last token that is not a padding
    token in each row. If no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since
    it cannot guess the padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take
    the last value in each row of the batch).
    c                   óH  ^ • \ rS rSrU 4S jr\         SS\R                  S-  S\R                  S-  S\R                  S-  S\R                  S-  S\R                  S-  S	\R                  S-  S
\	S-  S\	S-  S\	S-  S\
\R                     \-  4S jj5       rSrU =r$ )Ú"OpenAIGPTForSequenceClassificationi]  c                 óä   >• [         TU ]  U5        UR                  U l        [        U5      U l        [
        R                  " UR                  U R                  SS9U l        U R                  5         g r  )
r#   r$   r¥   rË   rÅ   r   r§   rw   Úscorerç   r  s     €r<   r$   Ú+OpenAIGPTForSequenceClassification.__init__i  sR   ø€ Ü‰Ñ˜Ô Ø ×+Ñ+ˆŒÜ)¨&Ó1ˆÔÜ—Y’Y˜vŸ}™}¨d¯o©oÀEÑJˆŒ
ð 	�‰Õr>   Nrò   rN   ró   rÌ   rô   r  rO   rõ   rö   r°   c
                 óÊ  • U	b  U	OU R                   R                  n	U R                  UUUUUUUU	S9nUS   nU R                  U5      nUb  UR                  SS u  pïOUR                  SS u  pïU R                   R
                  c  US:w  a  [        S5      eU R                   R
                  c  SnOÁUb�  XR                   R
                  :g  R                  UR                  [        R                  5      n[        R                  " UR                  S   UR                  [        R                  S9nUU-  R                  S5      nO.Sn[        R                  U R                  R                    S	35        U[        R                  " XíR                  S
9U4   nSnUGb�  U R                   R"                  c‘  U R$                  S:X  a  SU R                   l        OoU R$                  S:”  aN  UR&                  [        R(                  :X  d  UR&                  [        R*                  :X  a  SU R                   l        OSU R                   l        U R                   R"                  S:X  aJ  [-        5       nU R$                  S:X  a&  U" UR/                  5       UR/                  5       5      nOŽU" UU5      nO„U R                   R"                  S:X  a=  [1        5       nU" UR3                  SU R$                  5      UR3                  S5      5      nO-U R                   R"                  S:X  a  [5        5       nU" UU5      nU	(       d  U4USS -   nUb  U4U-   $ U$ [7        UUUR8                  UR:                  S9$ )ae  
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
    Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
    config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
    `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
Nr  r   rV   r   z=Cannot handle batch sizes > 1 if no padding token is defined.r@   )Údevicerµ   zŠ will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`)rA  Ú
regressionÚsingle_label_classificationÚmulti_label_classificationr  )r9   rö   rÅ   r>  r·   Úpad_token_idr'   rþ   rA  r)   Úint32rÍ   ÚargmaxÚloggerÚwarning_oncer;   rk   Úproblem_typer¥   rµ   r¸   r  r	   r¼   r   r,   r   r   r®   rÚ   )r7   rò   rN   ró   rÌ   rô   r  rO   rõ   rö   r  r"  r®   rØ   Ú
batch_sizeÚsequence_lengthÚlast_non_pad_tokenÚnon_pad_maskÚtoken_indicesÚpooled_logitsrÖ   r7  r½   s                          r<   rg   Ú*OpenAIGPTForSequenceClassification.forwardr  s  € ð( &1Ñ%<‘kÀ$Ç+Á+×BYÑBYˆà"×.Ñ.ØØ)Ø)Ø%Ø'Ø/Ø!5Ø#ð /ð 	
Ðð ,¨AÑ.ˆØ—‘˜MÓ*ˆàÑ Ø*3¯/©/¸"¸1Ð*=Ñ'ˆJ˜à*7×*=Ñ*=¸b¸qÐ*AÑ'ˆJð �;‰;×#Ñ#Ñ+°
¸a³ÜÐ\Ó]Ð]Ø�;‰;×#Ñ#Ñ+Ø!#ÑØÑ"à%¯©×)AÑ)AÑA×EÑEÀfÇmÁmÔUZ×U`ÑU`ÓaˆLÜ!ŸLšL¨¯©¸Ñ)<ÀVÇ]Á]ÔZ_×ZeÑZeÑfˆMØ"/°,Ñ">×!FÑ!FÀrÓ!JÑà!#ÐÜ×ÑØ—>‘>×*Ñ*Ð+ð ,Zð Zôð
 œuŸ|š|¨J¿}¹}ÑMÐOaÐaÑbˆàˆØÒØ�{‰{×'Ñ'Ñ/Ø—?‘? aÓ'Ø/;�D—K‘KÕ,Ø—_‘_ qÓ(¨f¯l©l¼e¿j¹jÓ.HÈFÏLÉLÔ\a×\eÑ\eÓLeØ/L�D—K‘KÕ,à/K�D—K‘KÔ,à�{‰{×'Ñ'¨<Ó7Ü"›9�Ø—?‘? aÓ'Ù# M×$9Ñ$9Ó$;¸V¿^¹^Ó=MÓN‘Dá# M°6Ó:‘DØ—‘×)Ñ)Ð-JÓJÜ+Ó-�Ù × 2Ñ 2°2°t·±Ó GÈÏÉÐUWËÓY‘Ø—‘×)Ñ)Ð-IÓIÜ,Ó.�Ù ¨vÓ6�ÞØ#Ð%Ð(;¸A¸BÐ(?Ñ?ˆFØ)-Ñ)9�T�G˜fÑ$ÐE¸vÐEä'ØØ Ø-×;Ñ;Ø*×5Ñ5ñ	
ð 	
r>   )r¥   r>  rÅ   )	NNNNNNNNN)rk   rl   rm   rn   r$   r   r)   rÁ   rÀ   r  rÛ   r  r   rg   ro   rp   rq   s   @r<   r<  r<  ]  s  ø† õð ð .2Ø37Ø26Ø04Ø26Ø*.Ø)-Ø,0Ø#'ñZ
à×#Ñ# dÑ*ðZ
ð ×)Ñ)¨DÑ0ðZ
ð ×(Ñ(¨4Ñ/ð	Z
ð
 ×&Ñ&¨Ñ-ðZ
ð ×(Ñ(¨4Ñ/ðZ
ð × Ñ  4Ñ'ðZ
ð   $™;ðZ
ð # T™kðZ
ð ˜D‘[ðZ
ð 
ˆu�|‰|Ñ	Ð7Ñ	7ôZ
ó öZ
r>   r<  )r.  r<  r  rË   rÃ   )6r¿   rE   Úcollections.abcr   Údataclassesr   Útypingr   r)   r   Útorch.nnr   r   r	   Ú r   rÉ   Úactivationsr   r   r   Ú
generationr   Úmodeling_outputsr   r   r   Úmodeling_utilsr   Úpytorch_utilsr   Úutilsr   r   r   Úconfiguration_openair   Ú
get_loggerrk   rH  ÚReLUry   ÚModuler   rs   r…   r–   rÃ   rÕ   rË   r  r.  r<  Ú__all__rÐ   r>   r<   Ú<module>rb     s«  ðñ  ã Ý $Ý !Ý ã Ý ß AÑ Aå &ß 9Ñ 9Ý )ß YÑ YÝ -Ý #÷ñ õ
 2ð 
×	Ò	˜HÓ	%€ð —7’7“9 d°HÀtÑ
L€ôC�—	‘	ô CôL ˆ"�)‰)ô  ôˆB�I‰Iô ô4`˜rŸy™yô `ðF ôU˜ó Uó ðUñ ðñð
 ô7 kó 7ó óð7ð( ôj
Ð-ó j
ó ðj
ñZ ðñôJÐ3°_ó JóðJñZ ðñôg
Ð 8ó g
óðg
ñT ðñ
ôe
Ð)Aó e
ó
ðe
òP�r>   