Ë
    S^(hA ã                   ó\  — d Z ddlZddlmZmZmZmZ ddlZddlZddlm	Z	 ddl
mZmZmZ ddlmZmZ ddlmZ dd	lmZmZmZmZmZmZmZmZ dd
lmZ ddlmZm Z m!Z! ddl"m#Z#m$Z$m%Z%m&Z&m'Z' ddl(m)Z)  e&jT                  e+«      Z,dZ-dZ.dZ/ G d„ de	j`                  «      Z1 G d„ de	j`                  «      Z2 G d„ de	j`                  «      Z3de2iZ4 G d„ de	j`                  «      Z5 G d„ de	j`                  «      Z6 G d„ de	j`                  «      Z7 G d„ d e	j`                  «      Z8 G d!„ d"e	j`                  «      Z9 G d#„ d$e	j`                  «      Z: G d%„ d&e«      Z;d'Z<d(Z= e$d)e<«       G d*„ d+e;«      «       Z> e$d,e<«       G d-„ d.e;e«      «       Z? e$d/e<«       G d0„ d1e;«      «       Z@ G d2„ d3e	j`                  «      ZA e$d4e<«       G d5„ d6e;«      «       ZB e$d7e<«       G d8„ d9e;«      «       ZC e$d:e<«       G d;„ d<e;«      «       ZD G d=„ d>e	j`                  «      ZE e$d?e<«       G d@„ dAe;«      «       ZFdDdB„ZGg dC¢ZHy)EzPyTorch Data2VecText model.é    N)ÚListÚOptionalÚTupleÚUnion)Únn)ÚBCEWithLogitsLossÚCrossEntropyLossÚMSELossé   )ÚACT2FNÚgelu)ÚGenerationMixin)Ú)BaseModelOutputWithPastAndCrossAttentionsÚ,BaseModelOutputWithPoolingAndCrossAttentionsÚ!CausalLMOutputWithCrossAttentionsÚMaskedLMOutputÚMultipleChoiceModelOutputÚQuestionAnsweringModelOutputÚSequenceClassifierOutputÚTokenClassifierOutput)ÚPreTrainedModel)Úapply_chunking_to_forwardÚ find_pruneable_heads_and_indicesÚprune_linear_layer)Úadd_code_sample_docstringsÚadd_start_docstringsÚ%add_start_docstrings_to_model_forwardÚloggingÚreplace_return_docstringsé   )ÚData2VecTextConfigé   zfacebook/data2vec-text-baser!   c                   ó2   ‡ — e Zd ZdZˆ fd„Z	 dd„Zd„ Zˆ xZS )ÚData2VecTextForTextEmbeddingszV
    Same as BertEmbeddings with a tiny tweak for positional embeddings indexing.
    c                 óÖ  •— t         ‰| �  «        t        j                  |j                  |j
                  |j                  ¬«      | _        t        j                  |j                  |j
                  «      | _	        t        j                  |j                  |j
                  «      | _        t        j                  |j
                  |j                  ¬«      | _        t        j                  |j                  «      | _        t#        |dd«      | _        | j'                  dt)        j*                  |j                  «      j-                  d«      d¬«       | j'                  d	t)        j.                  | j0                  j3                  «       t(        j4                  ¬
«      d¬«       |j                  | _        t        j                  |j                  |j
                  | j6                  ¬«      | _	        y )N)Úpadding_idx©ÚepsÚposition_embedding_typeÚabsoluteÚposition_ids)r    éÿÿÿÿF)Ú
persistentÚtoken_type_ids©Údtype)ÚsuperÚ__init__r   Ú	EmbeddingÚ
vocab_sizeÚhidden_sizeÚpad_token_idÚword_embeddingsÚmax_position_embeddingsÚposition_embeddingsÚtype_vocab_sizeÚtoken_type_embeddingsÚ	LayerNormÚlayer_norm_epsÚDropoutÚhidden_dropout_probÚdropoutÚgetattrr)   Úregister_bufferÚtorchÚarangeÚexpandÚzerosr+   ÚsizeÚlongr&   ©ÚselfÚconfigÚ	__class__s     €úq/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/transformers/models/data2vec/modeling_data2vec_text.pyr2   z&Data2VecTextForTextEmbeddings.__init__B   si  ø€ Ü‰ÑÔÜ!Ÿ|™|¨F×,=Ñ,=¸v×?QÑ?QÐ_e×_rÑ_rÔsˆÔÜ#%§<¡<°×0NÑ0NÐPV×PbÑPbÓ#cˆÔ Ü%'§\¡\°&×2HÑ2HÈ&×J\ÑJ\Ó%]ˆÔ"ô Ÿ™ f×&8Ñ&8¸f×>SÑ>SÔTˆŒÜ—z‘z &×"<Ñ"<Ó=ˆŒä'.¨vÐ7PÐR\Ó']ˆÔ$Ø×ÑØœEŸL™L¨×)GÑ)GÓH×OÑOÐPWÓXÐejð 	ô 	
ð 	×ÑØœeŸk™k¨$×*;Ñ*;×*@Ñ*@Ó*BÌ%Ï*É*ÔUÐbgð 	ô 	
ð
 "×.Ñ.ˆÔÜ#%§<¡<Ø×*Ñ*¨F×,>Ñ,>ÈD×L\ÑL\ô$
ˆÕ ó    c                 ó€  — |€+|�t        || j                  |«      }n| j                  |«      }|�|j                  «       }n|j                  «       d d }|d   }|€st	        | d«      r-| j
                  d d …d |…f   }|j                  |d   |«      }	|	}n:t        j                  |t        j                  | j                  j                  ¬«      }|€| j                  |«      }| j                  |«      }
||
z   }| j                  dk(  r| j                  |«      }||z  }| j!                  |«      }| j#                  |«      }|S )Nr,   r    r.   r   ©r0   Údevicer*   )Ú"create_position_ids_from_input_idsr&   Ú&create_position_ids_from_inputs_embedsrG   Úhasattrr.   rE   rC   rF   rH   r+   rQ   r7   r;   r)   r9   r<   r@   )rJ   Ú	input_idsr.   r+   Úinputs_embedsÚpast_key_values_lengthÚinput_shapeÚ
seq_lengthÚbuffered_token_type_idsÚ buffered_token_type_ids_expandedr;   Ú
embeddingsr9   s                rM   Úforwardz%Data2VecTextForTextEmbeddings.forward[   sR  € ð ÐØÐ$äAÀ)ÈT×M]ÑM]Ð_uÓv‘à#×JÑJÈ=ÓY�àÐ Ø#Ÿ.™.Ó*‰Kà'×,Ñ,Ó.¨s°Ð3ˆKà  ‘^ˆ
ð
 Ð!Ü�tÐ-Ô.Ø*.×*=Ñ*=ºaÀÀ*À¸nÑ*MÐ'Ø3J×3QÑ3QÐR]Ð^_ÑR`ÐblÓ3mÐ0Ø!A‘ä!&§¡¨[ÄÇ
Á
ÐSW×SdÑSd×SkÑSkÔ!l�àÐ Ø ×0Ñ0°Ó;ˆMØ $× :Ñ :¸>Ó JÐà"Ð%:Ñ:ˆ
Ø×'Ñ'¨:Ò5Ø"&×":Ñ":¸<Ó"HÐØÐ-Ñ-ˆJØ—^‘^ JÓ/ˆ
Ø—\‘\ *Ó-ˆ
ØÐrN   c                 ó  — |j                  «       dd }|d   }t        j                  | j                  dz   || j                  z   dz   t        j                  |j
                  ¬«      }|j                  d«      j                  |«      S )z×
        We are provided embeddings directly. We cannot infer which are padded so just generate sequential position ids.

        Args:
            inputs_embeds: torch.Tensor

        Returns: torch.Tensor
        Nr,   r    rP   r   )rG   rC   rD   r&   rH   rQ   Ú	unsqueezerE   )rJ   rV   rX   Úsequence_lengthr+   s        rM   rS   zDData2VecTextForTextEmbeddings.create_position_ids_from_inputs_embedsƒ   s€   € ð $×(Ñ(Ó*¨3¨BÐ/ˆØ% a™.ˆä—|‘|Ø×Ñ˜qÑ  /°D×4DÑ4DÑ"DÀqÑ"HÔPU×PZÑPZÐcp×cwÑcwô
ˆð ×%Ñ% aÓ(×/Ñ/°Ó<Ð<rN   )NNNNr   )Ú__name__Ú
__module__Ú__qualname__Ú__doc__r2   r]   rS   Ú__classcell__©rL   s   @rM   r$   r$   <   s   ø„ ñô

ð4 rsó&öP=rN   r$   c                   óP  ‡ — e Zd Zdˆ fd„	Zdej
                  dej
                  fd„Z	 	 	 	 	 	 ddej
                  deej                     deej                     deej                     d	eej                     d
ee	e	ej                           dee
   de	ej
                     fd„Zˆ xZS )ÚData2VecTextSelfAttentionc                 óâ  •— t         ‰| �  «        |j                  |j                  z  dk7  r2t	        |d«      s&t        d|j                  › d|j                  › d�«      ‚|j                  | _        t        |j                  |j                  z  «      | _        | j                  | j                  z  | _        t        j                  |j                  | j                  «      | _        t        j                  |j                  | j                  «      | _        t        j                  |j                  | j                  «      | _        t        j                  |j                  «      | _        |xs t#        |dd«      | _        | j$                  dk(  s| j$                  d	k(  rF|j&                  | _        t        j(                  d
|j&                  z  dz
  | j                  «      | _        |j,                  | _        y )Nr   Úembedding_sizezThe hidden size (z6) is not a multiple of the number of attention heads (ú)r)   r*   Úrelative_keyÚrelative_key_queryr"   r    )r1   r2   r5   Únum_attention_headsrT   Ú
ValueErrorÚintÚattention_head_sizeÚall_head_sizer   ÚLinearÚqueryÚkeyÚvaluer>   Úattention_probs_dropout_probr@   rA   r)   r8   r3   Údistance_embeddingÚ
is_decoder©rJ   rK   r)   rL   s      €rM   r2   z"Data2VecTextSelfAttention.__init__—   s�  ø€ Ü‰ÑÔØ×Ñ × :Ñ :Ñ:¸aÒ?ÌÐPVÐXhÔHiÜØ# F×$6Ñ$6Ð#7ð 8Ø ×4Ñ4Ð5°Qð8óð ð
 $*×#=Ñ#=ˆÔ Ü#& v×'9Ñ'9¸F×<VÑ<VÑ'VÓ#WˆÔ Ø!×5Ñ5¸×8PÑ8PÑPˆÔä—Y‘Y˜v×1Ñ1°4×3EÑ3EÓFˆŒ
Ü—9‘9˜V×/Ñ/°×1CÑ1CÓDˆŒÜ—Y‘Y˜v×1Ñ1°4×3EÑ3EÓFˆŒ
ä—z‘z &×"EÑ"EÓFˆŒØ'>ò (
Ä'ØÐ-¨zóC
ˆÔ$ð ×'Ñ'¨>Ò9¸T×=YÑ=YÐ]qÒ=qØ+1×+IÑ+IˆDÔ(Ü&(§l¡l°1°v×7UÑ7UÑ3UÐXYÑ3YÐ[_×[sÑ[sÓ&tˆDÔ#à ×+Ñ+ˆ�rN   ÚxÚreturnc                 ó¤   — |j                  «       d d | j                  | j                  fz   }|j                  |«      }|j	                  dddd«      S )Nr,   r   r"   r    r   )rG   rn   rq   ÚviewÚpermute)rJ   r{   Únew_x_shapes      rM   Útranspose_for_scoresz.Data2VecTextSelfAttention.transpose_for_scores±   sL   € Ø—f‘f“h˜s �m t×'?Ñ'?À×AYÑAYÐ&ZÑZˆØ�F‰F�;ÓˆØ�y‰y˜˜A˜q !Ó$Ð$rN   Úhidden_statesÚattention_maskÚ	head_maskÚencoder_hidden_statesÚencoder_attention_maskÚpast_key_valueÚoutput_attentionsc                 ó$  — | j                  |«      }|d u}	|	r|�|d   }
|d   }|}�n |	rC| j                  | j                  |«      «      }
| j                  | j                  |«      «      }|}n»|�y| j                  | j                  |«      «      }
| j                  | j                  |«      «      }t	        j
                  |d   |
gd¬«      }
t	        j
                  |d   |gd¬«      }n@| j                  | j                  |«      «      }
| j                  | j                  |«      «      }| j                  |«      }|d u}| j                  r|
|f}t	        j                  ||
j                  dd«      «      }| j                  dk(  s| j                  dk(  �r—|j                  d   |
j                  d   }}|rDt	        j                  |dz
  t        j                  |j                  ¬	«      j                  dd«      }n@t	        j                  |t        j                  |j                  ¬	«      j                  dd«      }t	        j                  |t        j                  |j                  ¬	«      j                  dd«      }||z
  }| j!                  || j"                  z   dz
  «      }|j%                  |j&                  ¬
«      }| j                  dk(  rt	        j(                  d||«      }||z   }nE| j                  dk(  r6t	        j(                  d||«      }t	        j(                  d|
|«      }||z   |z   }|t+        j,                  | j.                  «      z  }|�||z   }t0        j2                  j5                  |d¬«      }| j7                  |«      }|�||z  }t	        j                  ||«      }|j9                  dddd«      j;                  «       }|j=                  «       d d | j>                  fz   }|j                  |«      }|r||fn|f}| j                  r||fz   }|S )Nr   r    r"   ©Údimr,   éþÿÿÿrl   rm   rP   r/   zbhld,lrd->bhlrzbhrd,lrd->bhlrr   ) rt   r�   ru   rv   rC   Úcatry   ÚmatmulÚ	transposer)   ÚshapeÚtensorrH   rQ   r~   rD   rx   r8   Útor0   ÚeinsumÚmathÚsqrtrq   r   Ú
functionalÚsoftmaxr@   r   Ú
contiguousrG   rr   )rJ   r‚   rƒ   r„   r…   r†   r‡   rˆ   Úmixed_query_layerÚis_cross_attentionÚ	key_layerÚvalue_layerÚquery_layerÚ	use_cacheÚattention_scoresÚquery_lengthÚ
key_lengthÚposition_ids_lÚposition_ids_rÚdistanceÚpositional_embeddingÚrelative_position_scoresÚrelative_position_scores_queryÚrelative_position_scores_keyÚattention_probsÚcontext_layerÚnew_context_layer_shapeÚoutputss                               rM   r]   z!Data2VecTextSelfAttention.forward¶   sç  € ð !ŸJ™J }Ó5Ðð
 3¸$Ð>Ðá .Ð"<à& qÑ)ˆIØ(¨Ñ+ˆKØ3ŠNÙØ×1Ñ1°$·(±(Ð;PÓ2QÓRˆIØ×3Ñ3°D·J±JÐ?TÓ4UÓVˆKØ3‰NØÐ'Ø×1Ñ1°$·(±(¸=Ó2IÓJˆIØ×3Ñ3°D·J±J¸}Ó4MÓNˆKÜŸ	™	 >°!Ñ#4°iÐ"@ÀaÔHˆIÜŸ)™) ^°AÑ%6¸Ð$DÈ!ÔL‰Kà×1Ñ1°$·(±(¸=Ó2IÓJˆIØ×3Ñ3°D·J±J¸}Ó4MÓNˆKà×/Ñ/Ð0AÓBˆà"¨$Ð.ˆ	Ø�?Š?ð (¨Ð5ˆNô !Ÿ<™<¨°Y×5HÑ5HÈÈRÓ5PÓQÐà×'Ñ'¨>Ò9¸T×=YÑ=YÐ]qÓ=qØ'2×'8Ñ'8¸Ñ';¸Y¿_¹_ÈQÑ=O˜*ˆLÙÜ!&§¡¨j¸1©nÄEÇJÁJÐWd×WkÑWkÔ!l×!qÑ!qØ˜ó"‘ô "'§¡¨lÄ%Ç*Á*ÐUb×UiÑUiÔ!j×!oÑ!oÐprÐtuÓ!v�Ü"Ÿ\™\¨*¼E¿J¹JÈ}×OcÑOcÔd×iÑiÐjkÐmoÓpˆNØ%¨Ñ6ˆHà#'×#:Ñ#:¸8Àd×FbÑFbÑ;bÐefÑ;fÓ#gÐ Ø#7×#:Ñ#:À×ARÑARÐ#:Ó#SÐ à×+Ñ+¨~Ò=Ü+0¯<©<Ð8HÈ+ÐWkÓ+lÐ(Ø#3Ð6NÑ#NÑ Ø×-Ñ-Ð1EÒEÜ16·±Ð>NÐP[Ð]qÓ1rÐ.Ü/4¯|©|Ð<LÈiÐYmÓ/nÐ,Ø#3Ð6TÑ#TÐWsÑ#sÐ à+¬d¯i©i¸×8PÑ8PÓ.QÑQÐØÐ%à/°.Ñ@Ðô Ÿ-™-×/Ñ/Ð0@ÀbÐ/ÓIˆð Ÿ,™, Ó7ˆð Ð Ø-°	Ñ9ˆOäŸ™ _°kÓBˆà%×-Ñ-¨a°°A°qÓ9×DÑDÓFˆØ"/×"4Ñ"4Ó"6°s¸Ð";¸t×?QÑ?QÐ>SÑ"SÐØ%×*Ñ*Ð+BÓCˆá6G�= /Ñ2ÈmÐM]ˆà�?Š?Ø Ð 1Ñ1ˆGØˆrN   ©N©NNNNNF)ra   rb   rc   r2   rC   ÚTensorr�   r   ÚFloatTensorr   Úboolr]   re   rf   s   @rM   rh   rh   –   så   ø„ õ,ð4% e§l¡lð %°u·|±|ó %ð 7;Ø15Ø=AØ>BØDHØ,1ñcà—|‘|ðcð ! ×!2Ñ!2Ñ3ðcð ˜E×-Ñ-Ñ.ð	cð
  (¨×(9Ñ(9Ñ:ðcð !)¨×):Ñ):Ñ ;ðcð !  u¨U×->Ñ->Ñ'?Ñ!@ÑAðcð $ D™>ðcð 
ˆu�|‰|Ñ	÷crN   rh   c                   ón   ‡ — e Zd Zˆ fd„Zdej
                  dej
                  dej
                  fd„Zˆ xZS )ÚData2VecTextSelfOutputc                 ó(  •— t         ‰| �  «        t        j                  |j                  |j                  «      | _        t        j                  |j                  |j                  ¬«      | _        t        j                  |j                  «      | _
        y ©Nr'   )r1   r2   r   rs   r5   Údenser<   r=   r>   r?   r@   rI   s     €rM   r2   zData2VecTextSelfOutput.__init__  s`   ø€ Ü‰ÑÔÜ—Y‘Y˜v×1Ñ1°6×3EÑ3EÓFˆŒ
ÜŸ™ f×&8Ñ&8¸f×>SÑ>SÔTˆŒÜ—z‘z &×"<Ñ"<Ó=ˆ�rN   r‚   Úinput_tensorr|   c                 ór   — | j                  |«      }| j                  |«      }| j                  ||z   «      }|S r­   ©r¶   r@   r<   ©rJ   r‚   r·   s      rM   r]   zData2VecTextSelfOutput.forward$  ó7   € ØŸ
™
 =Ó1ˆØŸ™ ]Ó3ˆØŸ™ }°|Ñ'CÓDˆØÐrN   ©ra   rb   rc   r2   rC   r¯   r]   re   rf   s   @rM   r³   r³     ó1   ø„ ô>ð U§\¡\ð ÀÇÁð ÐRW×R^ÑR^÷ rN   r³   Úeagerc                   ó  ‡ — e Zd Zdˆ fd„	Zd„ Z	 	 	 	 	 	 ddej                  deej                     deej                     deej                     deej                     dee	e	ej                           d	ee
   d
e	ej                     fd„Zˆ xZS )ÚData2VecTextAttentionc                 óž   •— t         ‰| �  «        t        |j                     ||¬«      | _        t        |«      | _        t        «       | _        y )N©r)   )	r1   r2   Ú$DATA2VEC_TEXT_SELF_ATTENTION_CLASSESÚ_attn_implementationrJ   r³   ÚoutputÚsetÚpruned_headsrz   s      €rM   r2   zData2VecTextAttention.__init__2  sC   ø€ Ü‰ÑÔÜ8¸×9TÑ9TÑUØÐ,Cô
ˆŒ	ô -¨VÓ4ˆŒÜ›EˆÕrN   c                 ó>  — t        |«      dk(  ry t        || j                  j                  | j                  j                  | j
                  «      \  }}t        | j                  j                  |«      | j                  _        t        | j                  j                  |«      | j                  _        t        | j                  j                  |«      | j                  _	        t        | j                  j                  |d¬«      | j                  _        | j                  j                  t        |«      z
  | j                  _        | j                  j                  | j                  j                  z  | j                  _        | j
                  j                  |«      | _        y )Nr   r    rŠ   )Úlenr   rJ   rn   rq   rÇ   r   rt   ru   rv   rÅ   r¶   rr   Úunion)rJ   ÚheadsÚindexs      rM   Úprune_headsz!Data2VecTextAttention.prune_heads:  s  € Üˆu‹:˜Š?ØÜ7Ø�4—9‘9×0Ñ0°$·)±)×2OÑ2OÐQU×QbÑQbó
‰ˆˆuô
 -¨T¯Y©Y¯_©_¸eÓDˆ�	‰	ŒÜ*¨4¯9©9¯=©=¸%Ó@ˆ�	‰	ŒÜ,¨T¯Y©Y¯_©_¸eÓDˆ�	‰	ŒÜ.¨t¯{©{×/@Ñ/@À%ÈQÔOˆ�‰Ôð )-¯	©	×(EÑ(EÌÈEË
Ñ(Rˆ�	‰	Ô%Ø"&§)¡)×"?Ñ"?À$Ç)Á)×B_ÑB_Ñ"_ˆ�	‰	ÔØ ×-Ñ-×3Ñ3°EÓ:ˆÕrN   r‚   rƒ   r„   r…   r†   r‡   rˆ   r|   c           	      óp   — | j                  |||||||«      }| j                  |d   |«      }	|	f|dd  z   }
|
S )Nr   r    )rJ   rÅ   )rJ   r‚   rƒ   r„   r…   r†   r‡   rˆ   Úself_outputsÚattention_outputr¬   s              rM   r]   zData2VecTextAttention.forwardL  sW   € ð —y‘yØØØØ!Ø"ØØó
ˆð  Ÿ;™; |°A¡¸ÓFÐØ#Ð%¨°Q°RÐ(8Ñ8ˆØˆrN   r­   r®   )ra   rb   rc   r2   rÍ   rC   r¯   r   r°   r   r±   r]   re   rf   s   @rM   rÀ   rÀ   1  sÆ   ø„ õ"ò;ð* 7;Ø15Ø=AØ>BØDHØ,1ñà—|‘|ðð ! ×!2Ñ!2Ñ3ðð ˜E×-Ñ-Ñ.ð	ð
  (¨×(9Ñ(9Ñ:ðð !)¨×):Ñ):Ñ ;ðð !  u¨U×->Ñ->Ñ'?Ñ!@ÑAðð $ D™>ðð 
ˆu�|‰|Ñ	÷rN   rÀ   c                   óV   ‡ — e Zd Zˆ fd„Zdej
                  dej
                  fd„Zˆ xZS )ÚData2VecTextIntermediatec                 ó  •— t         ‰| �  «        t        j                  |j                  |j
                  «      | _        t        |j                  t        «      rt        |j                     | _        y |j                  | _        y r­   )r1   r2   r   rs   r5   Úintermediate_sizer¶   Ú
isinstanceÚ
hidden_actÚstrr   Úintermediate_act_fnrI   s     €rM   r2   z!Data2VecTextIntermediate.__init__f  s]   ø€ Ü‰ÑÔÜ—Y‘Y˜v×1Ñ1°6×3KÑ3KÓLˆŒ
Ü�f×'Ñ'¬Ô-Ü'-¨f×.?Ñ.?Ñ'@ˆDÕ$à'-×'8Ñ'8ˆDÕ$rN   r‚   r|   c                 óJ   — | j                  |«      }| j                  |«      }|S r­   )r¶   rØ   )rJ   r‚   s     rM   r]   z Data2VecTextIntermediate.forwardn  s&   € ØŸ
™
 =Ó1ˆØ×0Ñ0°Ó?ˆØÐrN   r¼   rf   s   @rM   rÒ   rÒ   e  s#   ø„ ô9ð U§\¡\ð °e·l±l÷ rN   rÒ   c                   ón   ‡ — e Zd Zˆ fd„Zdej
                  dej
                  dej
                  fd„Zˆ xZS )ÚData2VecTextOutputc                 ó(  •— t         ‰| �  «        t        j                  |j                  |j
                  «      | _        t        j                  |j
                  |j                  ¬«      | _        t        j                  |j                  «      | _        y rµ   )r1   r2   r   rs   rÔ   r5   r¶   r<   r=   r>   r?   r@   rI   s     €rM   r2   zData2VecTextOutput.__init__v  s`   ø€ Ü‰ÑÔÜ—Y‘Y˜v×7Ñ7¸×9KÑ9KÓLˆŒ
ÜŸ™ f×&8Ñ&8¸f×>SÑ>SÔTˆŒÜ—z‘z &×"<Ñ"<Ó=ˆ�rN   r‚   r·   r|   c                 ór   — | j                  |«      }| j                  |«      }| j                  ||z   «      }|S r­   r¹   rº   s      rM   r]   zData2VecTextOutput.forward|  r»   rN   r¼   rf   s   @rM   rÛ   rÛ   u  r½   rN   rÛ   c                   ó  ‡ — e Zd Zˆ fd„Z	 	 	 	 	 	 ddej
                  deej                     deej                     deej                     deej                     deeeej                           dee	   d	eej
                     fd
„Z
d„ Zˆ xZS )ÚData2VecTextLayerc                 óf  •— t         ‰| �  «        |j                  | _        d| _        t	        |«      | _        |j                  | _        |j                  | _        | j                  r,| j                  st        | › d�«      ‚t	        |d¬«      | _	        t        |«      | _        t        |«      | _        y )Nr    z> should be used as a decoder model if cross attention is addedr*   rÂ   )r1   r2   Úchunk_size_feed_forwardÚseq_len_dimrÀ   Ú	attentionry   Úadd_cross_attentionro   ÚcrossattentionrÒ   ÚintermediaterÛ   rÅ   rI   s     €rM   r2   zData2VecTextLayer.__init__…  s—   ø€ Ü‰ÑÔØ'-×'EÑ'EˆÔ$ØˆÔÜ.¨vÓ6ˆŒØ ×+Ñ+ˆŒØ#)×#=Ñ#=ˆÔ Ø×#Ò#Ø—?’?Ü  D 6Ð)gÐ!hÓiÐiÜ"7¸ÐXbÔ"cˆDÔÜ4°VÓ<ˆÔÜ(¨Ó0ˆ�rN   r‚   rƒ   r„   r…   r†   r‡   rˆ   r|   c           	      óÒ  — |�|d d nd }| j                  |||||¬«      }	|	d   }
| j                  r|	dd }|	d   }n|	dd  }d }| j                  rT|�Rt        | d«      st        d| › d�«      ‚|�|d	d  nd }| j	                  |
||||||«      }|d   }
||dd z   }|d   }|z   }t        | j                  | j                  | j                  |
«      }|f|z   }| j                  r|fz   }|S )
Nr"   )rˆ   r‡   r   r    r,   rå   z'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`rŒ   )	rã   ry   rT   ro   rå   r   Úfeed_forward_chunkrá   râ   )rJ   r‚   rƒ   r„   r…   r†   r‡   rˆ   Úself_attn_past_key_valueÚself_attention_outputsrÐ   r¬   Úpresent_key_valueÚcross_attn_present_key_valueÚcross_attn_past_key_valueÚcross_attention_outputsÚlayer_outputs                    rM   r]   zData2VecTextLayer.forward“  s}  € ð :HÐ9S >°"°1Ñ#5ÐY]Ð Ø!%§¡ØØØØ/Ø3ð "0ó "
Ðð 2°!Ñ4Ðð �?Š?Ø,¨Q¨rÐ2ˆGØ 6°rÑ :Ñà,¨Q¨RÐ0ˆGà'+Ð$Ø�?Š?Ð4Ð@Ü˜4Ð!1Ô2Ü Ø=¸d¸Vð DDð Dóð ð @NÐ?Y¨°r°sÑ(;Ð_cÐ%Ø&*×&9Ñ&9Ø ØØØ%Ø&Ø)Ø!ó'Ð#ð  7°qÑ9ÐØÐ 7¸¸"Ð =Ñ=ˆGð ,CÀ2Ñ+FÐ(Ø 1Ð4PÑ PÐä0Ø×#Ñ# T×%AÑ%AÀ4×CSÑCSÐUeó
ˆð  �/ GÑ+ˆð �?Š?ØÐ!2Ð 4Ñ4ˆGàˆrN   c                 óL   — | j                  |«      }| j                  ||«      }|S r­   )ræ   rÅ   )rJ   rÐ   Úintermediate_outputrï   s       rM   rè   z$Data2VecTextLayer.feed_forward_chunkÔ  s,   € Ø"×/Ñ/Ð0@ÓAÐØ—{‘{Ð#6Ð8HÓIˆØÐrN   r®   )ra   rb   rc   r2   rC   r¯   r   r°   r   r±   r]   rè   re   rf   s   @rM   rß   rß   „  sÇ   ø„ ô1ð" 7;Ø15Ø=AØ>BØDHØ,1ñ?à—|‘|ð?ð ! ×!2Ñ!2Ñ3ð?ð ˜E×-Ñ-Ñ.ð	?ð
  (¨×(9Ñ(9Ñ:ð?ð !)¨×):Ñ):Ñ ;ð?ð !  u¨U×->Ñ->Ñ'?Ñ!@ÑAð?ð $ D™>ð?ð 
ˆu�|‰|Ñ	ó?öBrN   rß   c                   óD  ‡ — e Zd Zˆ fd„Z	 	 	 	 	 	 	 	 	 ddej
                  deej                     deej                     deej                     deej                     deeeej                           dee	   d	ee	   d
ee	   dee	   de
eej
                     ef   fd„Zˆ xZS )ÚData2VecTextEncoderc                 óÐ   •— t         ‰| �  «        || _        t        j                  t        |j                  «      D �cg c]  }t        |«      ‘Œ c}«      | _        d| _	        y c c}w )NF)
r1   r2   rK   r   Ú
ModuleListÚrangeÚnum_hidden_layersrß   ÚlayerÚgradient_checkpointing)rJ   rK   Ú_rL   s      €rM   r2   zData2VecTextEncoder.__init__Ü  sO   ø€ Ü‰ÑÔØˆŒÜ—]‘]ÄuÈV×MeÑMeÓGfÖ#gÀ!Ô$5°fÕ$=Ò#gÓhˆŒ
Ø&+ˆÕ#ùò $hs   ½A#r‚   rƒ   r„   r…   r†   Úpast_key_valuesrž   rˆ   Úoutput_hidden_statesÚreturn_dictr|   c                 óš  — |	rdnd }|rdnd }|r| j                   j                  rdnd }| j                  r%| j                  r|rt        j                  d«       d}|rdnd }t        | j                  «      D ]¤  \  }}|	r||fz   }|�||   nd }|�||   nd }| j                  r/| j                  r#| j                  |j                  |||||||«      }n ||||||||«      }|d   }|r	||d   fz  }|sŒ|||d   fz   }| j                   j                  sŒœ||d   fz   }Œ¦ |	r||fz   }|
st        d„ |||||fD «       «      S t        |||||¬	«      S )
N© zZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...Fr   r,   r    r"   c              3   ó$   K  — | ]  }|�|–— Œ
 y ­wr­   rÿ   )Ú.0Úvs     rM   ú	<genexpr>z.Data2VecTextEncoder.forward.<locals>.<genexpr>$  s   è ø€ ò 
àð �=ô ñ
ùs   ‚)Úlast_hidden_staterû   r‚   Ú
attentionsÚcross_attentions)rK   rä   rù   ÚtrainingÚloggerÚwarning_onceÚ	enumeraterø   Ú_gradient_checkpointing_funcÚ__call__Útupler   )rJ   r‚   rƒ   r„   r…   r†   rû   rž   rˆ   rü   rý   Úall_hidden_statesÚall_self_attentionsÚall_cross_attentionsÚnext_decoder_cacheÚiÚlayer_moduleÚlayer_head_maskr‡   Úlayer_outputss                       rM   r]   zData2VecTextEncoder.forwardâ  sÎ  € ñ #7™B¸DÐÙ$5™b¸4ÐÙ%6¸4¿;¹;×;ZÒ;Z™rÐ`dÐà×&Ò&¨4¯=ª=ÙÜ×#Ñ#Øpôð "�	á#,™R°$ÐÜ(¨¯©Ó4ò #	V‰OˆAˆ|Ù#Ø$5¸Ð8HÑ$HÐ!à.7Ð.C˜i¨šlÈˆOØ3BÐ3N˜_¨QÒ/ÐTXˆNà×*Ò*¨t¯}ª}Ø $× AÑ AØ ×)Ñ)Ø!Ø"Ø#Ø)Ø*Ø"Ø%ó	!‘ñ !-Ø!Ø"Ø#Ø)Ø*Ø"Ø%ó!�ð *¨!Ñ,ˆMÙØ" }°RÑ'8Ð&:Ñ:Ð"Ú Ø&9¸]È1Ñ=MÐ<OÑ&OÐ#Ø—;‘;×2Ó2Ø+?À=ÐQRÑCSÐBUÑ+UÑ(ðG#	VñJ  Ø 1°]Ð4DÑ DÐáÜñ 
ð "Ø&Ø%Ø'Ø(ðô
ó 
ð 
ô 9Ø+Ø.Ø+Ø*Ø1ô
ð 	
rN   )	NNNNNNFFT)ra   rb   rc   r2   rC   r¯   r   r°   r   r±   r   r   r]   re   rf   s   @rM   ró   ró   Û  s  ø„ ô,ð 7;Ø15Ø=AØ>BØEIØ$(Ø,1Ø/4Ø&*ñS
à—|‘|ðS
ð ! ×!2Ñ!2Ñ3ðS
ð ˜E×-Ñ-Ñ.ð	S
ð
  (¨×(9Ñ(9Ñ:ðS
ð !)¨×):Ñ):Ñ ;ðS
ð " %¨¨e×.?Ñ.?Ñ(@Ñ"AÑBðS
ð ˜D‘>ðS
ð $ D™>ðS
ð ' t™nðS
ð ˜d‘^ðS
ð 
ˆu�U—\‘\Ñ"Ð$MÐMÑ	N÷S
rN   ró   c                   óV   ‡ — e Zd Zˆ fd„Zdej
                  dej
                  fd„Zˆ xZS )ÚData2VecTextPoolerc                 ó²   •— t         ‰| �  «        t        j                  |j                  |j                  «      | _        t        j                  «       | _        y r­   )r1   r2   r   rs   r5   r¶   ÚTanhÚ
activationrI   s     €rM   r2   zData2VecTextPooler.__init__:  s9   ø€ Ü‰ÑÔÜ—Y‘Y˜v×1Ñ1°6×3EÑ3EÓFˆŒ
ÜŸ'™'›)ˆ�rN   r‚   r|   c                 ó\   — |d d …df   }| j                  |«      }| j                  |«      }|S ©Nr   )r¶   r  )rJ   r‚   Úfirst_token_tensorÚpooled_outputs       rM   r]   zData2VecTextPooler.forward?  s6   € ð +ª1¨a¨4Ñ0ÐØŸ
™
Ð#5Ó6ˆØŸ™¨Ó6ˆØÐrN   r¼   rf   s   @rM   r  r  9  s#   ø„ ô$ð
 U§\¡\ð °e·l±l÷ rN   r  c                   ó*   — e Zd ZdZeZdZdZddgZd„ Z	y)ÚData2VecTextPreTrainedModelz†
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    Údata2vec_textTr$   rß   c                 óf  — t        |t        j                  «      rm|j                  j                  j                  d| j                  j                  ¬«       |j                  �%|j                  j                  j                  «        yyt        |t        j                  «      rz|j                  j                  j                  d| j                  j                  ¬«       |j                  �2|j                  j                  |j                     j                  «        yyt        |t        j                  «      r|t        |d«      r0|j                  �$|j                  j                  j                  «        t        |d«      r3|j                  �&|j                  j                  j                  d«       yyyy)zInitialize the weightsg        )ÚmeanÚstdNÚbiasÚweightg      ð?)rÕ   r   rs   r&  ÚdataÚnormal_rK   Úinitializer_ranger%  Úzero_r3   r&   r<   rT   Úfill_)rJ   Úmodules     rM   Ú_init_weightsz)Data2VecTextPreTrainedModel._init_weightsS  s5  € ä�fœbŸi™iÔ(ð �M‰M×Ñ×&Ñ&¨C°T·[±[×5RÑ5RÐ&ÔSØ�{‰{Ð&Ø—‘× Ñ ×&Ñ&Õ(ð 'ä˜¤§¡Ô-Ø�M‰M×Ñ×&Ñ&¨C°T·[±[×5RÑ5RÐ&ÔSØ×!Ñ!Ð-Ø—‘×"Ñ" 6×#5Ñ#5Ñ6×<Ñ<Õ>ð .ä˜¤§¡Ô-Ü�v˜vÔ&¨6¯;©;Ð+BØ—‘× Ñ ×&Ñ&Ô(Ü�v˜xÔ(¨V¯]©]Ð-FØ—‘×"Ñ"×(Ñ(¨Õ-ð .GÐ(ð .rN   N)
ra   rb   rc   rd   r!   Úconfig_classÚbase_model_prefixÚsupports_gradient_checkpointingÚ_no_split_modulesr-  rÿ   rN   rM   r   r   H  s+   „ ñð
 &€LØ'ÐØ&*Ð#Ø8Ð:MÐNÐó.rN   r   aB  
    Data2VecText was proposed in [data2vec: A General Framework for Self-supervised Learning in Speech, Vision and
    Language](https://arxiv.org/pdf/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu and
    Michael Auli.

    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 ([`Data2VecTextConfig`]): 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.
zoThe bare Data2VecText Model for text transformer outputting raw hidden-states without any specific head on top.c            !       ó  ‡ — e Zd ZdZdˆ fd„	Zd„ Zd„ Zd„ Z ee	j                  d«      «       eeee¬«      	 	 	 	 	 	 	 	 	 	 	 	 	 ddeej"                     d	eej"                     d
eej"                     deej"                     deej"                     deej"                     deej"                     deej"                     deeej&                        dee   dee   dee   dee   deeej"                     ef   fd„«       «       Zˆ xZS )ÚData2VecTextModela*  

    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 `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.

    .. _*Attention is all you need*: https://arxiv.org/abs/1706.03762

    c                 óº   •— t         ‰| �  |«       || _        t        |«      | _        t        |«      | _        |rt        |«      nd | _        | j                  «        y r­   )
r1   r2   rK   r$   r\   ró   Úencoderr  ÚpoolerÚ	post_init)rJ   rK   Úadd_pooling_layerrL   s      €rM   r2   zData2VecTextModel.__init__¿  sL   ø€ Ü‰Ñ˜Ô ØˆŒä7¸Ó?ˆŒÜ*¨6Ó2ˆŒá4EÔ(¨Ô0È4ˆŒð 	�‰ÕrN   c                 ó.   — | j                   j                  S r­   ©r\   r7   ©rJ   s    rM   Úget_input_embeddingsz&Data2VecTextModel.get_input_embeddingsË  s   € Ø�‰×.Ñ.Ð.rN   c                 ó&   — || j                   _        y r­   r:  )rJ   rv   s     rM   Úset_input_embeddingsz&Data2VecTextModel.set_input_embeddingsÎ  s   € Ø*/ˆ�‰Õ'rN   c                 ó˜   — |j                  «       D ]7  \  }}| j                  j                  |   j                  j	                  |«       Œ9 y)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)Úitemsr5  rø   rã   rÍ   )rJ   Úheads_to_prunerø   rË   s       rM   Ú_prune_headszData2VecTextModel._prune_headsÑ  sE   € ð
 +×0Ñ0Ó2ò 	C‰LˆE�5Ø�L‰L×Ñ˜uÑ%×/Ñ/×;Ñ;¸EÕBñ	CrN   úbatch_size, sequence_length©Ú
checkpointÚoutput_typer.  rU   rƒ   r.   r+   r„   rV   r…   r†   rû   rž   rˆ   rü   rý   r|   c                 óœ  — |�|n| j                   j                  }|�|n| j                   j                  }|�|n| j                   j                  }| j                   j                  r|
�|
n| j                   j
                  }
nd}
|�|�t        d«      ‚|�#| j                  ||«       |j                  «       }n!|�|j                  «       dd }nt        d«      ‚|\  }}|�|j                  n|j                  }|	�|	d   d   j                  d   nd}|€t        j                  |||z   f|¬«      }|€pt        | j                  d	«      r4| j                  j                  dd…d|…f   }|j!                  ||«      }|}n&t        j"                  |t        j$                  |¬
«      }| j'                  ||«      }| j                   j                  rE|�C|j                  «       \  }}}||f}|€t        j                  ||¬«      }| j)                  |«      }nd}| j+                  || j                   j,                  «      }| j                  |||||¬«      }| j/                  ||||||	|
|||¬«
      }|d   }| j0                  �| j1                  |«      nd}|s
||f|dd z   S t3        |||j4                  |j6                  |j8                  |j:                  ¬«      S )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 timer,   z5You have to specify either input_ids or inputs_embedsr   r"   )rQ   r.   rP   )rU   r+   r.   rV   rW   )	rƒ   r„   r…   r†   rû   rž   rˆ   rü   rý   r    )r  Úpooler_outputrû   r‚   r  r  )rK   rˆ   rü   Úuse_return_dictry   rž   ro   Ú%warn_if_padding_and_no_attention_maskrG   rQ   r�   rC   ÚonesrT   r\   r.   rE   rF   rH   Úget_extended_attention_maskÚinvert_attention_maskÚget_head_maskr÷   r5  r6  r   rû   r‚   r  r  )rJ   rU   rƒ   r.   r+   r„   rV   r…   r†   rû   rž   rˆ   rü   rý   rX   Ú
batch_sizerY   rQ   rW   rZ   r[   Úextended_attention_maskÚencoder_batch_sizeÚencoder_sequence_lengthrú   Úencoder_hidden_shapeÚencoder_extended_attention_maskÚembedding_outputÚencoder_outputsÚsequence_outputr  s                                  rM   r]   zData2VecTextModel.forwardÙ  s  € ðV 2CÐ1NÑ-ÐTX×T_ÑT_×TqÑTqÐà$8Ð$DÑ È$Ï+É+×JjÑJjð 	ð &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà�;‰;×!Ò!Ø%.Ð%:™	ÀÇÁ×@UÑ@U‰IàˆIàÐ  ]Ð%>ÜÐcÓdÐdØÐ"Ø×6Ñ6°yÀ.ÔQØ#Ÿ.™.Ó*‰KØÐ&Ø'×,Ñ,Ó.¨s°Ð3‰KäÐTÓUÐUà!,Ñˆ
�JØ%.Ð%:�×!Ò!À×@TÑ@Tˆð DSÐC^ °Ñ!3°AÑ!6×!<Ñ!<¸QÒ!?ÐdeÐàÐ!Ü"ŸZ™Z¨*°jÐCYÑ6YÐ)ZÐdjÔkˆNàÐ!Ü�t—‘Ð(8Ô9Ø*.¯/©/×*HÑ*HÊÈKÈZÈKÈÑ*XÐ'Ø3J×3QÑ3QÐR\Ð^hÓ3iÐ0Ø!A‘ä!&§¡¨[ÄÇ
Á
ÐSYÔ!Z�ð 15×0PÑ0PÐQ_ÐalÓ0mÐð �;‰;×!Ò!Ð&;Ð&GØ=R×=WÑ=WÓ=YÑ:ÐÐ 7¸Ø$6Ð8OÐ#PÐ Ø%Ð-Ü).¯©Ð4HÐQWÔ)XÐ&Ø.2×.HÑ.HÐI_Ó.`Ñ+à.2Ð+ð ×&Ñ& y°$·+±+×2OÑ2OÓPˆ	àŸ?™?ØØ%Ø)Ø'Ø#9ð +ó 
Ðð Ÿ,™,ØØ2ØØ"7Ø#BØ+ØØ/Ø!5Ø#ð 'ó 
ˆð *¨!Ñ,ˆØ8<¿¹Ð8O˜Ÿ™ OÔ4ÐUYˆáØ# ]Ð3°oÀaÀbÐ6IÑIÐIä;Ø-Ø'Ø+×;Ñ;Ø)×7Ñ7Ø&×1Ñ1Ø,×=Ñ=ô
ð 	
rN   )T)NNNNNNNNNNNNN)ra   rb   rc   rd   r2   r<  r>  rB  r   ÚDATA2VECTEXT_INPUTS_DOCSTRINGÚformatr   Ú_CHECKPOINT_FOR_DOCr   Ú_CONFIG_FOR_DOCr   rC   r¯   r   r°   r±   r   r   r]   re   rf   s   @rM   r3  r3  «  s™  ø„ ñ
õ
ò/ò0òCñ +Ð+H×+OÑ+OÐPmÓ+nÓoÙØ&Ø@Ø$ôð -1Ø15Ø15Ø/3Ø,0Ø04Ø8<Ø9=Ø=AØ$(Ø,0Ø/3Ø&*ñ@
à˜EŸL™LÑ)ð@
ð ! §¡Ñ.ð@
ð ! §¡Ñ.ð	@
ð
 ˜uŸ|™|Ñ,ð@
ð ˜EŸL™LÑ)ð@
ð   §¡Ñ-ð@
ð  (¨¯©Ñ5ð@
ð !)¨¯©Ñ 6ð@
ð " $ u×'8Ñ'8Ñ"9Ñ:ð@
ð ˜D‘>ð@
ð $ D™>ð@
ð ' t™nð@
ð ˜d‘^ð@
ð 
ˆu�U—\‘\Ñ"Ð$PÐPÑ	Qò@
óó pô@
rN   r3  zNData2VecText Model with a `language modeling` head on top for CLM fine-tuning.c            #       ó  ‡ — e Zd ZddgZˆ fd„Zd„ Zd„ Z eej                  d«      «       e
ee¬«      	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddeej                     d	eej                      d
eej                     deej                     deej                      deej                      deej                      deej                      deej                     deeeej                            dee   dee   dee   dee   deeef   fd„«       «       Zd„ Zˆ xZS )ÚData2VecTextForCausalLMúlm_head.decoder.weightúlm_head.decoder.biasc                 óÊ   •— t         ‰| �  |«       |j                  st        j	                  d«       t        |d¬«      | _        t        |«      | _        | j                  «        y )NzTIf you want to use `Data2VecTextLMHeadModel` as a standalone, add `is_decoder=True.`F©r8  ©
r1   r2   ry   r  Úwarningr3  r!  ÚData2VecTextLMHeadÚlm_headr7  rI   s     €rM   r2   z Data2VecTextForCausalLM.__init__i  sM   ø€ Ü‰Ñ˜Ô à× Ò Ü�N‰NÐqÔrä.¨vÈÔOˆÔÜ)¨&Ó1ˆŒð 	�‰ÕrN   c                 ó.   — | j                   j                  S r­   ©re  Údecoderr;  s    rM   Úget_output_embeddingsz-Data2VecTextForCausalLM.get_output_embeddingsu  ó   € Ø�|‰|×#Ñ#Ð#rN   c                 ó&   — || j                   _        y r­   rg  ©rJ   Únew_embeddingss     rM   Úset_output_embeddingsz-Data2VecTextForCausalLM.set_output_embeddingsx  ó   € Ø-ˆ�‰ÕrN   rC  )rF  r.  rU   rƒ   r.   r+   r„   rV   r…   r†   Úlabelsrû   rž   rˆ   rü   rý   r|   c                 óž  — |�|n| j                   j                  }|	�d}| j                  |||||||||
||||¬«      }|d   }| j                  |«      }d}|	�* | j                  ||	fd| j                   j
                  i|¤Ž}|s|f|dd z   }|�|f|z   S |S t        |||j                  |j                  |j                  |j                  ¬«      S )aX
  
        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**.

        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the left-to-right language modeling loss (next word prediction). 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]`
        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`).

        Returns:

        Example:

        ```python
        >>> from transformers import AutoTokenizer, Data2VecTextForCausalLM, Data2VecTextConfig
        >>> import torch

        >>> tokenizer = AutoTokenizer.from_pretrained("facebook/data2vec-text-base")
        >>> config = Data2VecTextConfig.from_pretrained("facebook/data2vec-text-base")
        >>> config.is_decoder = True
        >>> model = Data2VecTextForCausalLM.from_pretrained("facebook/data2vec-text-base", config=config)

        >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
        >>> outputs = model(**inputs)

        >>> prediction_logits = outputs.logits
        ```NF)rƒ   r.   r+   r„   rV   r…   r†   rû   rž   rˆ   rü   rý   r   r4   r"   )ÚlossÚlogitsrû   r‚   r  r  )rK   rI  r!  re  Úloss_functionr4   r   rû   r‚   r  r  )rJ   rU   rƒ   r.   r+   r„   rV   r…   r†   rp  rû   rž   rˆ   rü   rý   Úkwargsr¬   rW  Úprediction_scoresÚlm_lossrÅ   s                        rM   r]   zData2VecTextForCausalLM.forward{  s#  € ð~ &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆØÐØˆIà×$Ñ$ØØ)Ø)Ø%ØØ'Ø"7Ø#9Ø+ØØ/Ø!5Ø#ð %ó 
ˆð  " !™*ˆØ ŸL™L¨Ó9ÐàˆØÐØ(�d×(Ñ(Ø!Øñð  Ÿ;™;×1Ñ1ðð ñ	ˆGñ Ø'Ð)¨G°A°B¨KÑ7ˆFØ,3Ð,?�W�J Ñ'ÐKÀVÐKä0ØØ$Ø#×3Ñ3Ø!×/Ñ/Ø×)Ñ)Ø$×5Ñ5ô
ð 	
rN   c                 óJ   ‡— d}|D ]  }|t        ˆfd„|D «       «      fz  }Œ |S )Nrÿ   c              3   ót   •K  — | ]/  }|j                  d ‰j                  |j                  «      «      –— Œ1 y­w)r   N)Úindex_selectr’   rQ   )r  Ú
past_stateÚbeam_idxs     €rM   r  z9Data2VecTextForCausalLM._reorder_cache.<locals>.<genexpr>ë  s.   øè ø€ ÒnÐU_�j×-Ñ-¨a°·±¸Z×=NÑ=NÓ1O×PÑnùs   ƒ58)r  )rJ   rû   r|  Úreordered_pastÚ
layer_pasts     `  rM   Ú_reorder_cachez&Data2VecTextForCausalLM._reorder_cacheç  s=   ø€ ØˆØ)ò 	ˆJØÜÓnÐcmÔnÓnðñ ‰Nð	ð ÐrN   )NNNNNNNNNNNNNN)ra   rb   rc   Ú_tied_weights_keysr2   ri  rn  r   rX  rY  r   r   r[  r   rC   Ú
LongTensorr°   r   r±   r   r]   r  re   rf   s   @rM   r]  r]  c  sÃ  ø„ ð 3Ð4JÐKÐô
ò$ò.ñ +Ð+H×+OÑ+OÐPmÓ+nÓoÙÐ+LÐ[jÔkð 15Ø6:Ø59Ø37Ø15Ø59Ø=AØ>BØ-1ØEIØ$(Ø,0Ø/3Ø&*ñh
à˜E×,Ñ,Ñ-ðh
ð ! ×!2Ñ!2Ñ3ðh
ð ! ×!1Ñ!1Ñ2ð	h
ð
 ˜u×/Ñ/Ñ0ðh
ð ˜E×-Ñ-Ñ.ðh
ð   × 1Ñ 1Ñ2ðh
ð  (¨×(9Ñ(9Ñ:ðh
ð !)¨×):Ñ):Ñ ;ðh
ð ˜×)Ñ)Ñ*ðh
ð " %¨¨e×.?Ñ.?Ñ(@Ñ"AÑBðh
ð ˜D‘>ðh
ð $ D™>ðh
ð ' t™nðh
ð ˜d‘^ðh
ð" 
ˆuÐ7Ð7Ñ	8ò#h
ó ló pðh
öTrN   r]  z6data2vec Model with a `language modeling` head on top.c                   óà  ‡ — e Zd ZddgZˆ fd„Zd„ Zd„ Z eej                  d«      «       e
eeed¬«      	 	 	 	 	 	 	 	 	 	 	 	 dd	eej                      d
eej"                     deej                      deej                      deej"                     deej"                     deej"                     deej"                     deej                      dee   dee   dee   deeef   fd„«       «       Zˆ xZS )ÚData2VecTextForMaskedLMr^  r_  c                 óÊ   •— t         ‰| �  |«       |j                  rt        j	                  d«       t        |d¬«      | _        t        |«      | _        | j                  «        y )NzsIf you want to use `Data2VecTextForMaskedLM` make sure `config.is_decoder=False` for bi-directional self-attention.Fra  rb  rI   s     €rM   r2   z Data2VecTextForMaskedLM.__init__ô  sS   ø€ Ü‰Ñ˜Ô à×ÒÜ�N‰Nð1ôô
 /¨vÈÔOˆÔÜ)¨&Ó1ˆŒð 	�‰ÕrN   c                 ó.   — | j                   j                  S r­   rg  r;  s    rM   ri  z-Data2VecTextForMaskedLM.get_output_embeddings  rj  rN   c                 ó&   — || j                   _        y r­   rg  rl  s     rM   rn  z-Data2VecTextForMaskedLM.set_output_embeddings  ro  rN   rC  z<mask>)rE  rF  r.  ÚmaskrU   rƒ   r.   r+   r„   rV   r…   r†   rp  rˆ   rü   rý   r|   c                 óÔ  — |�|n| j                   j                  }| j                  |||||||||
||¬«      }|d   }| j                  |«      }d}|	�at	        «       }|	j                  |j                  «      }	 ||j                  d| j                   j                  «      |	j                  d«      «      }|s|f|dd z   }|�|f|z   S |S t        |||j                  |j                  ¬«      S )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]`
        kwargs (`Dict[str, any]`, *optional*, defaults to *{}*):
            Used to hide legacy arguments that have been deprecated.
        N)
rƒ   r.   r+   r„   rV   r…   r†   rˆ   rü   rý   r   r,   r"   ©rr  rs  r‚   r  )rK   rI  r!  re  r	   r’   rQ   r~   r4   r   r‚   r  )rJ   rU   rƒ   r.   r+   r„   rV   r…   r†   rp  rˆ   rü   rý   r¬   rW  rv  Úmasked_lm_lossÚloss_fctrÅ   s                      rM   r]   zData2VecTextForMaskedLM.forward	  s  € ð< &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà×$Ñ$ØØ)Ø)Ø%ØØ'Ø"7Ø#9Ø/Ø!5Ø#ð %ó 
ˆð " !™*ˆØ ŸL™L¨Ó9ÐàˆØÐÜ'Ó)ˆHà—Y‘YÐ0×7Ñ7Ó8ˆFÙ%Ð&7×&<Ñ&<¸RÀÇÁ×AWÑAWÓ&XÐZ`×ZeÑZeÐfhÓZiÓjˆNáØ'Ð)¨G°A°B¨KÑ7ˆFØ3AÐ3M�^Ð%¨Ñ.ÐYÐSYÐYäØØ$Ø!×/Ñ/Ø×)Ñ)ô	
ð 	
rN   )NNNNNNNNNNNN)ra   rb   rc   r€  r2   ri  rn  r   rX  rY  r   rZ  r   r[  r   rC   r�  r°   r±   r   r   r]   re   rf   s   @rM   rƒ  rƒ  ð  s}  ø„ à2Ð4JÐKÐôò$ò.ñ +Ð+H×+OÑ+OÐPmÓ+nÓoÙØ&Ø"Ø$Øô	ð 15Ø6:Ø59Ø37Ø15Ø59Ø=AØ>BØ-1Ø,0Ø/3Ø&*ñ9
à˜E×,Ñ,Ñ-ð9
ð ! ×!2Ñ!2Ñ3ð9
ð ! ×!1Ñ!1Ñ2ð	9
ð
 ˜u×/Ñ/Ñ0ð9
ð ˜E×-Ñ-Ñ.ð9
ð   × 1Ñ 1Ñ2ð9
ð  (¨×(9Ñ(9Ñ:ð9
ð !)¨×):Ñ):Ñ ;ð9
ð ˜×)Ñ)Ñ*ð9
ð $ D™>ð9
ð ' t™nð9
ð ˜d‘^ð9
ð 
ˆu�nÐ$Ñ	%ò9
óó pô9
rN   rƒ  c                   ó.   ‡ — e Zd ZdZˆ fd„Zd„ Zd„ Zˆ xZS )rd  z/Data2VecText Head for masked language modeling.c                 óâ  •— t         ‰| �  «        t        j                  |j                  |j                  «      | _        t        j                  |j                  |j                  ¬«      | _        t        j                  |j                  |j                  «      | _
        t        j                  t        j                  |j                  «      «      | _        | j                  | j                  _        y rµ   )r1   r2   r   rs   r5   r¶   r<   r=   Ú
layer_normr4   rh  Ú	ParameterrC   rF   r%  rI   s     €rM   r2   zData2VecTextLMHead.__init__P  s—   ø€ Ü‰ÑÔÜ—Y‘Y˜v×1Ñ1°6×3EÑ3EÓFˆŒ
ÜŸ,™, v×'9Ñ'9¸v×?TÑ?TÔUˆŒä—y‘y ×!3Ñ!3°V×5FÑ5FÓGˆŒÜ—L‘L¤§¡¨V×->Ñ->Ó!?Ó@ˆŒ	Ø ŸI™Iˆ�‰ÕrN   c                 ó‚   — | j                  |«      }t        |«      }| j                  |«      }| j                  |«      }|S r­   )r¶   r   rŽ  rh  ©rJ   Úfeaturesru  r{   s       rM   r]   zData2VecTextLMHead.forwardY  s;   € Ø�J‰J�xÓ ˆÜ�‹GˆØ�O‰O˜AÓˆð �L‰L˜‹OˆàˆrN   c                 óÌ   — | j                   j                  j                  j                  dk(  r| j                  | j                   _        y | j                   j                  | _        y )NÚmeta)rh  r%  rQ   Útyper;  s    rM   Ú_tie_weightszData2VecTextLMHead._tie_weightsc  sC   € ð �<‰<×Ñ×#Ñ#×(Ñ(¨FÒ2Ø $§	¡	ˆD�L‰LÕàŸ™×)Ñ)ˆD�IrN   )ra   rb   rc   rd   r2   r]   r–  re   rf   s   @rM   rd  rd  M  s   ø„ Ù9ô&òö*rN   rd  z¤
    Data2VecText Model transformer with a sequence classification/regression head on top (a linear layer on top of the
    pooled output) e.g. for GLUE tasks.
    c                   óŠ  ‡ — e Zd Zˆ fd„Z eej                  d«      «       eee	e
¬«      	 	 	 	 	 	 	 	 	 	 ddeej                     deej                     deej                     deej                     deej                     d	eej                     d
eej                     dee   dee   dee   deee	f   fd„«       «       Zˆ xZS )Ú%Data2VecTextForSequenceClassificationc                 ó¸   •— t         ‰| �  |«       |j                  | _        || _        t	        |d¬«      | _        t        |«      | _        | j                  «        y ©NFra  )	r1   r2   Ú
num_labelsrK   r3  r!  ÚData2VecTextClassificationHeadÚ
classifierr7  rI   s     €rM   r2   z.Data2VecTextForSequenceClassification.__init__t  sK   ø€ Ü‰Ñ˜Ô Ø ×+Ñ+ˆŒØˆŒä.¨vÈÔOˆÔÜ8¸Ó@ˆŒð 	�‰ÕrN   rC  rD  rU   rƒ   r.   r+   r„   rV   rp  rˆ   rü   rý   r|   c                 óT  — |
�|
n| j                   j                  }
| j                  ||||||||	|
¬«	      }|d   }| j                  |«      }d}|��¢|j	                  |j
                  «      }| j                   j                  €�| j                  dk(  rd| j                   _        nl| j                  dkD  rL|j                  t        j                  k(  s|j                  t        j                  k(  rd| j                   _        nd| j                   _        | j                   j                  dk(  rIt        «       }| j                  dk(  r& ||j                  «       |j                  «       «      }nŒ |||«      }n‚| j                   j                  dk(  r=t        «       } ||j                  d| j                  «      |j                  d«      «      }n,| j                   j                  dk(  rt!        «       } |||«      }|
s|f|d	d z   }|�|f|z   S |S t#        |||j$                  |j&                  ¬
«      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).
        N©rƒ   r.   r+   r„   rV   rˆ   rü   rý   r   r    Ú
regressionÚsingle_label_classificationÚmulti_label_classificationr,   r"   r‰  )rK   rI  r!  r�  r’   rQ   Úproblem_typer›  r0   rC   rH   rp   r
   Úsqueezer	   r~   r   r   r‚   r  ©rJ   rU   rƒ   r.   r+   r„   rV   rp  rˆ   rü   rý   r¬   rW  rs  rr  r‹  rÅ   s                    rM   r]   z-Data2VecTextForSequenceClassification.forward  sî  € ð2 &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà×$Ñ$ØØ)Ø)Ø%ØØ'Ø/Ø!5Ø#ð %ó 

ˆð " !™*ˆØ—‘ Ó1ˆàˆØÑØ—Y‘Y˜vŸ}™}Ó-ˆFà�{‰{×'Ñ'Ð/Ø—?‘? aÒ'Ø/;�D—K‘KÕ,Ø—_‘_ qÒ(¨f¯l©l¼e¿j¹jÒ.HÈFÏLÉLÔ\a×\eÑ\eÒLeØ/L�D—K‘KÕ,à/K�D—K‘KÔ,à�{‰{×'Ñ'¨<Ò7Ü"›9�Ø—?‘? aÒ'Ù# F§N¡NÓ$4°f·n±nÓ6FÓG‘Dá# F¨FÓ3‘DØ—‘×)Ñ)Ð-JÒJÜ+Ó-�Ù §¡¨B°·±Ó @À&Ç+Á+ÈbÃ/ÓR‘Ø—‘×)Ñ)Ð-IÒIÜ,Ó.�Ù ¨Ó/�áØ�Y ¨¨ Ñ,ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä'ØØØ!×/Ñ/Ø×)Ñ)ô	
ð 	
rN   ©
NNNNNNNNNN)ra   rb   rc   r2   r   rX  rY  r   rZ  r   r[  r   rC   r�  r°   r±   r   r   r]   re   rf   s   @rM   r˜  r˜  l  sB  ø„ ô	ñ +Ð+H×+OÑ+OÐPmÓ+nÓoÙØ&Ø,Ø$ôð 15Ø6:Ø59Ø37Ø15Ø59Ø-1Ø,0Ø/3Ø&*ñE
à˜E×,Ñ,Ñ-ðE
ð ! ×!2Ñ!2Ñ3ðE
ð ! ×!1Ñ!1Ñ2ð	E
ð
 ˜u×/Ñ/Ñ0ðE
ð ˜E×-Ñ-Ñ.ðE
ð   × 1Ñ 1Ñ2ðE
ð ˜×)Ñ)Ñ*ðE
ð $ D™>ðE
ð ' t™nðE
ð ˜d‘^ðE
ð 
ˆuÐ.Ð.Ñ	/òE
óó pôE
rN   r˜  z­
    Data2VecText 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.
    c                   óŠ  ‡ — e Zd Zˆ fd„Z eej                  d«      «       eee	e
¬«      	 	 	 	 	 	 	 	 	 	 ddeej                     deej                     deej                     deej                     deej                     d	eej                     d
eej                     dee   dee   dee   deee	f   fd„«       «       Zˆ xZS )ÚData2VecTextForMultipleChoicec                 óö   •— t         ‰| �  |«       t        |«      | _        t	        j
                  |j                  «      | _        t	        j                  |j                  d«      | _
        | j                  «        y )Nr    )r1   r2   r3  r!  r   r>   r?   r@   rs   r5   r�  r7  rI   s     €rM   r2   z&Data2VecTextForMultipleChoice.__init__Õ  sW   ø€ Ü‰Ñ˜Ô ä.¨vÓ6ˆÔÜ—z‘z &×"<Ñ"<Ó=ˆŒÜŸ)™) F×$6Ñ$6¸Ó:ˆŒð 	�‰ÕrN   z(batch_size, num_choices, sequence_lengthrD  rU   r.   rƒ   rp  r+   r„   rV   rˆ   rü   rý   r|   c                 ó‚  — |
�|
n| j                   j                  }
|�|j                  d   n|j                  d   }|�!|j                  d|j	                  d«      «      nd}|�!|j                  d|j	                  d«      «      nd}|�!|j                  d|j	                  d«      «      nd}|�!|j                  d|j	                  d«      «      nd}|�1|j                  d|j	                  d«      |j	                  d«      «      nd}| j                  ||||||||	|
¬«	      }|d   }| j                  |«      }| j                  |«      }|j                  d|«      }d}|�.t        «       }|j                  |j                  «      } |||«      }|
s|f|dd z   }|�|f|z   S |S t        |||j                  |j                  ¬«      S )aJ  
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
            num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
            `input_ids` above)
        Nr    r,   rŒ   )r+   r.   rƒ   r„   rV   rˆ   rü   rý   r"   r‰  )rK   rI  r�   r~   rG   r!  r@   r�  r	   r’   rQ   r   r‚   r  )rJ   rU   r.   rƒ   rp  r+   r„   rV   rˆ   rü   rý   Únum_choicesÚflat_input_idsÚflat_position_idsÚflat_token_type_idsÚflat_attention_maskÚflat_inputs_embedsr¬   r  rs  Úreshaped_logitsrr  r‹  rÅ   s                           rM   r]   z%Data2VecTextForMultipleChoice.forwardß  sö  € ð6 &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆØ,5Ð,A�i—o‘o aÒ(À}×GZÑGZÐ[\ÑG]ˆàCLÐCX˜Ÿ™¨¨I¯N©N¸2Ó,>Ô?Ð^bˆØLXÐLd˜L×-Ñ-¨b°,×2CÑ2CÀBÓ2GÔHÐjnÐØR`ÐRl˜n×1Ñ1°"°n×6IÑ6IÈ"Ó6MÔNÐrvÐØR`ÐRl˜n×1Ñ1°"°n×6IÑ6IÈ"Ó6MÔNÐrvÐð Ð(ð ×Ñ˜r =×#5Ñ#5°bÓ#9¸=×;MÑ;MÈbÓ;QÔRàð 	ð ×$Ñ$ØØ*Ø.Ø.ØØ,Ø/Ø!5Ø#ð %ó 

ˆð   ™
ˆàŸ™ ]Ó3ˆØ—‘ Ó/ˆØ Ÿ+™+ b¨+Ó6ˆàˆØÐÜ'Ó)ˆHà—Y‘Y˜×5Ñ5Ó6ˆFÙ˜O¨VÓ4ˆDáØ%Ð'¨'°!°"¨+Ñ5ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä(ØØ"Ø!×/Ñ/Ø×)Ñ)ô	
ð 	
rN   r¦  )ra   rb   rc   r2   r   rX  rY  r   rZ  r   r[  r   rC   r�  r°   r±   r   r   r]   re   rf   s   @rM   r¨  r¨  Í  sE  ø„ ôñ +Ø%×,Ñ,Ð-WÓXóñ  Ø&Ø-Ø$ôð 15Ø59Ø6:Ø-1Ø37Ø15Ø59Ø,0Ø/3Ø&*ñA
à˜E×,Ñ,Ñ-ðA
ð ! ×!1Ñ!1Ñ2ðA
ð ! ×!2Ñ!2Ñ3ð	A
ð
 ˜×)Ñ)Ñ*ðA
ð ˜u×/Ñ/Ñ0ðA
ð ˜E×-Ñ-Ñ.ðA
ð   × 1Ñ 1Ñ2ðA
ð $ D™>ðA
ð ' t™nðA
ð ˜d‘^ðA
ð 
ˆuÐ/Ð/Ñ	0òA
óóôA
rN   r¨  z«
    Data2VecText 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.
    c                   óŠ  ‡ — e Zd Zˆ fd„Z eej                  d«      «       eee	e
¬«      	 	 	 	 	 	 	 	 	 	 ddeej                     deej                     deej                     deej                     deej                     d	eej                     d
eej                     dee   dee   dee   deee	f   fd„«       «       Zˆ xZS )Ú"Data2VecTextForTokenClassificationc                 ód  •— t         ‰| �  |«       |j                  | _        t        |d¬«      | _        |j
                  �|j
                  n|j                  }t        j                  |«      | _	        t        j                  |j                  |j                  «      | _        | j                  «        y rš  )r1   r2   r›  r3  r!  Úclassifier_dropoutr?   r   r>   r@   rs   r5   r�  r7  ©rJ   rK   rµ  rL   s      €rM   r2   z+Data2VecTextForTokenClassification.__init__3  s‹   ø€ Ü‰Ñ˜Ô Ø ×+Ñ+ˆŒä.¨vÈÔOˆÔà)/×)BÑ)BÐ)NˆF×%Ò%ÐTZ×TnÑTnð 	ô —z‘zÐ"4Ó5ˆŒÜŸ)™) F×$6Ñ$6¸×8IÑ8IÓJˆŒð 	�‰ÕrN   rC  rD  rU   rƒ   r.   r+   r„   rV   rp  rˆ   rü   rý   r|   c                 óÞ  — |
�|
n| j                   j                  }
| j                  ||||||||	|
¬«	      }|d   }| j                  |«      }| j	                  |«      }d}|�Wt        «       }|j                  |j                  «      } ||j                  d| j                  «      |j                  d«      «      }|
s|f|dd z   }|�|f|z   S |S t        |||j                  |j                  ¬«      S )zÛ
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
        NrŸ  r   r,   r"   r‰  )rK   rI  r!  r@   r�  r	   r’   rQ   r~   r›  r   r‚   r  r¥  s                    rM   r]   z*Data2VecTextForTokenClassification.forwardA  s
  € ð. &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà×$Ñ$ØØ)Ø)Ø%ØØ'Ø/Ø!5Ø#ð %ó 

ˆð " !™*ˆàŸ,™, Ó7ˆØ—‘ Ó1ˆàˆØÐÜ'Ó)ˆHà—Y‘Y˜vŸ}™}Ó-ˆFÙ˜FŸK™K¨¨D¯O©OÓ<¸f¿k¹kÈ"»oÓNˆDáØ�Y ¨¨ Ñ,ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä$ØØØ!×/Ñ/Ø×)Ñ)ô	
ð 	
rN   r¦  )ra   rb   rc   r2   r   rX  rY  r   rZ  r   r[  r   rC   r�  r°   r±   r   r   r]   re   rf   s   @rM   r³  r³  +  s5  ø„ ôñ +Ð+H×+OÑ+OÐPmÓ+nÓoÙØ&Ø)Ø$ôð 15Ø6:Ø59Ø37Ø15Ø59Ø-1Ø,0Ø/3Ø&*ñ4
à˜E×,Ñ,Ñ-ð4
ð ! ×!2Ñ!2Ñ3ð4
ð ! ×!1Ñ!1Ñ2ð	4
ð
 ˜u×/Ñ/Ñ0ð4
ð ˜E×-Ñ-Ñ.ð4
ð   × 1Ñ 1Ñ2ð4
ð ˜×)Ñ)Ñ*ð4
ð $ D™>ð4
ð ' t™nð4
ð ˜d‘^ð4
ð 
ˆuÐ+Ð+Ñ	,ò4
óó pô4
rN   r³  c                   ó(   ‡ — e Zd ZdZˆ fd„Zd„ Zˆ xZS )rœ  z-Head for sentence-level classification tasks.c                 óZ  •— t         ‰| �  «        t        j                  |j                  |j                  «      | _        |j                  �|j                  n|j                  }t        j                  |«      | _	        t        j                  |j                  |j                  «      | _        y r­   )r1   r2   r   rs   r5   r¶   rµ  r?   r>   r@   r›  Úout_projr¶  s      €rM   r2   z'Data2VecTextClassificationHead.__init__‚  s   ø€ Ü‰ÑÔÜ—Y‘Y˜v×1Ñ1°6×3EÑ3EÓFˆŒ
à)/×)BÑ)BÐ)NˆF×%Ò%ÐTZ×TnÑTnð 	ô —z‘zÐ"4Ó5ˆŒÜŸ	™	 &×"4Ñ"4°f×6GÑ6GÓHˆ�rN   c                 óÐ   — |d d …dd d …f   }| j                  |«      }| j                  |«      }t        j                  |«      }| j                  |«      }| j	                  |«      }|S r  )r@   r¶   rC   Útanhrº  r‘  s       rM   r]   z&Data2VecTextClassificationHead.forward‹  sY   € Ø’Q˜š1�WÑˆØ�L‰L˜‹OˆØ�J‰J�q‹MˆÜ�J‰J�q‹MˆØ�L‰L˜‹OˆØ�M‰M˜!ÓˆØˆrN   )ra   rb   rc   rd   r2   r]   re   rf   s   @rM   rœ  rœ    s   ø„ Ù7ôIörN   rœ  zå
    Data2VecText 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`).
    c                   óª  ‡ — e Zd Zˆ fd„Z eej                  d«      «       eee	e
¬«      	 	 	 	 	 	 	 	 	 	 	 ddeej                     deej                     deej                     deej                     deej                     d	eej                     d
eej                     deej                     dee   dee   dee   deee	f   fd„«       «       Zˆ xZS )Ú Data2VecTextForQuestionAnsweringc                 óè   •— t         ‰| �  |«       |j                  | _        t        |d¬«      | _        t        j                  |j                  |j                  «      | _        | j                  «        y rš  )
r1   r2   r›  r3  r!  r   rs   r5   Ú
qa_outputsr7  rI   s     €rM   r2   z)Data2VecTextForQuestionAnswering.__init__�  sV   ø€ Ü‰Ñ˜Ô Ø ×+Ñ+ˆŒä.¨vÈÔOˆÔÜŸ)™) F×$6Ñ$6¸×8IÑ8IÓJˆŒð 	�‰ÕrN   rC  rD  rU   rƒ   r.   r+   r„   rV   Ústart_positionsÚend_positionsrˆ   rü   rý   r|   c                 ó(  — |�|n| j                   j                  }| j                  |||||||	|
|¬«	      }|d   }| j                  |«      }|j	                  dd¬«      \  }}|j                  d«      j                  «       }|j                  d«      j                  «       }d}|�·|�µt        |j                  «       «      dkD  r|j                  d«      }t        |j                  «       «      dkD  r|j                  d«      }|j                  d«      }|j                  d|«      }|j                  d|«      }t        |¬«      } |||«      } |||«      }||z   dz  }|s||f|dd z   }|�|f|z   S |S t        ||||j                  |j                  ¬	«      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.
        NrŸ  r   r    r,   rŠ   )Úignore_indexr"   )rr  Ústart_logitsÚ
end_logitsr‚   r  )rK   rI  r!  rÀ  Úsplitr¤  r˜   rÉ   rG   Úclampr	   r   r‚   r  )rJ   rU   rƒ   r.   r+   r„   rV   rÁ  rÂ  rˆ   rü   rý   r¬   rW  rs  rÅ  rÆ  Ú
total_lossÚignored_indexr‹  Ú
start_lossÚend_lossrÅ   s                          rM   r]   z(Data2VecTextForQuestionAnswering.forward§  sÄ  € ð< &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà×$Ñ$ØØ)Ø)Ø%ØØ'Ø/Ø!5Ø#ð %ó 

ˆð " !™*ˆà—‘ Ó1ˆØ#)§<¡<°°r <Ó#:Ñ ˆ�jØ#×+Ñ+¨BÓ/×:Ñ:Ó<ˆØ×'Ñ'¨Ó+×6Ñ6Ó8ˆ
àˆ
ØÐ&¨=Ð+Dä�?×'Ñ'Ó)Ó*¨QÒ.Ø"1×"9Ñ"9¸"Ó"=�Ü�=×%Ñ%Ó'Ó(¨1Ò,Ø -× 5Ñ 5°bÓ 9�à(×-Ñ-¨aÓ0ˆMØ-×3Ñ3°A°}ÓEˆOØ)×/Ñ/°°=ÓAˆMä'°]ÔCˆHÙ! ,°Ó@ˆJÙ 
¨MÓ:ˆHØ$ xÑ/°1Ñ4ˆJáØ" JÐ/°'¸!¸"°+Ñ=ˆFØ/9Ð/E�Z�M FÑ*ÐQÈ6ÐQä+ØØ%Ø!Ø!×/Ñ/Ø×)Ñ)ô
ð 	
rN   )NNNNNNNNNNN)ra   rb   rc   r2   r   rX  rY  r   rZ  r   r[  r   rC   r�  r°   r±   r   r   r]   re   rf   s   @rM   r¾  r¾  •  s[  ø„ ôñ +Ð+H×+OÑ+OÐPmÓ+nÓoÙØ&Ø0Ø$ôð 15Ø6:Ø59Ø37Ø15Ø59Ø6:Ø48Ø,0Ø/3Ø&*ñH
à˜E×,Ñ,Ñ-ðH
ð ! ×!2Ñ!2Ñ3ðH
ð ! ×!1Ñ!1Ñ2ð	H
ð
 ˜u×/Ñ/Ñ0ðH
ð ˜E×-Ñ-Ñ.ðH
ð   × 1Ñ 1Ñ2ðH
ð " %×"2Ñ"2Ñ3ðH
ð   × 0Ñ 0Ñ1ðH
ð $ D™>ðH
ð ' t™nðH
ð ˜d‘^ðH
ð 
ˆuÐ2Ð2Ñ	3òH
óó pôH
rN   r¾  c                 ó¾   — | j                  |«      j                  «       }t        j                  |d¬«      j	                  |«      |z   |z  }|j                  «       |z   S )a  
    Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols
    are ignored. This is modified from fairseq's `utils.make_positions`.

    Args:
        x: torch.Tensor x:

    Returns: torch.Tensor
    r    rŠ   )Únerp   rC   ÚcumsumÚtype_asrH   )rU   r&   rW   r‡  Úincremental_indicess        rM   rR   rR   ø  sW   € ð �<‰<˜Ó$×(Ñ(Ó*€DÜ Ÿ<™<¨°!Ô4×<Ñ<¸TÓBÐE[Ñ[Ð_cÑcÐØ×#Ñ#Ó%¨Ñ3Ð3rN   )r]  rƒ  r¨  r¾  r˜  r³  r3  r   )r   )Ird   r”   Útypingr   r   r   r   rC   Útorch.utils.checkpointr   Útorch.nnr   r	   r
   Úactivationsr   r   Ú
generationr   Úmodeling_outputsr   r   r   r   r   r   r   r   Úmodeling_utilsr   Úpytorch_utilsr   r   r   Úutilsr   r   r   r   r   Úconfiguration_data2vec_textr!   Ú
get_loggerra   r  Ú_HIDDEN_STATES_START_POSITIONrZ  r[  ÚModuler$   rh   r³   rÃ   rÀ   rÒ   rÛ   rß   ró   r  r   ÚDATA2VECTEXT_START_DOCSTRINGrX  r3  r]  rƒ  rd  r˜  r¨  r³  rœ  r¾  rR   Ú__all__rÿ   rN   rM   ú<module>rá     s®  ðñ "ã ß /Ó /ã Û Ý ß AÑ Aç 'Ý )÷	÷ 	ó 	õ .ß lÑ l÷õ õ <ð 
ˆ×	Ñ	˜HÓ	%€ð !"Ð ð 4Ð Ø&€ôV= B§I¡Iô V=ôtC §	¡	ô CôN˜RŸY™Yô ð Ð&ð(Ð $ô0˜BŸI™Iô 0ôh˜rŸy™yô ô ˜Ÿ™ô ôS˜Ÿ	™	ô SônZ
˜"Ÿ)™)ô Z
ô|˜Ÿ™ô ô. /ô .ð< Ð ð&/!Ð ñd ØuØ óôq
Ð3ó q
ó	ðq
ñh ØXÐZvóôGÐ9¸?ó GóðGñT ÐRÐTpÓqôX
Ð9ó X
ó rðX
ôx*˜Ÿ™ô *ñ> ðð !óôW
Ð,Gó W
óðW
ñt ðð !óôT
Ð$?ó T
óðT
ñn ðð !óôI
Ð)Dó I
óðI
ôZ R§Y¡Yô ñ, ðð !óôY
Ð'Bó Y
óðY
óx4ò 	�rN   