Ë
    S^(hµ‹  ã                   óø  — d Z ddl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 ddl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 ddlmZ  ej@                  e!«      Z"dZ#d„ Z$d„ Z%d%d„Z& G d„ dejN                  «      Z(d„ Z) G d„ dejN                  «      Z* G d„ de«      Z+dZ,dZ- ede,«       G d„ de+«      «       Z. ede,«       G d„ d e+e«      «       Z/ ed!e,«       G d"„ d#e+«      «       Z0g d$¢Z1y)&zPyTorch CTRL model.é    )ÚOptionalÚTupleÚUnionN)Únn)ÚBCEWithLogitsLossÚCrossEntropyLossÚMSELossé   )ÚGenerationMixin)ÚBaseModelOutputWithPastÚCausalLMOutputWithPastÚSequenceClassifierOutput)ÚPreTrainedModel)ÚConv1DÚ find_pruneable_heads_and_indicesÚprune_linear_layer)Úadd_start_docstringsÚ%add_start_docstrings_to_model_forwardÚloggingÚreplace_return_docstringsé   )Ú
CTRLConfigr   c                 óP   — dt        j                  dd|dz  z  |z  «      z  }| |z  S )Nr   i'  é   )ÚtorchÚpow)ÚposÚiÚd_model_sizeÚangle_ratess       úd/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/transformers/models/ctrl/modeling_ctrl.pyÚ
angle_defnr"   &   s/   € Ø”e—i‘i ¨¨Q°!©V©¸Ñ'DÓEÑE€KØ�ÑÐó    c                 óÒ  — t        t        j                  | t        j                  ¬«      j	                  |«      j                  d«      t        j                  |t        j                  ¬«      j	                  |«      j                  d«      |«      }t        j                  |d d …dd d…f   «      }t        j                  |d d …dd d…f   «      }t        j                  ||gd¬«      }|S )N©Údtyper   r   r   éÿÿÿÿ©Údim)	r"   r   ÚarangeÚint64ÚtoÚ	unsqueezeÚsinÚcosÚcat)Úpositionr   r&   Ú
angle_radsÚsinesÚcosinesÚpos_encodings          r!   Úpositional_encodingr6   +   s¶   € äÜ�‰�X¤U§[¡[Ô1×4Ñ4°UÓ;×EÑEÀaÓHÜ�‰�\¬¯©Ô5×8Ñ8¸Ó?×IÑIÈ!ÓLØó€Jô �I‰I�j¢ A D q D Ñ)Ó*€EÜ�i‰i˜
¢1 a d¨ d 7Ñ+Ó,€Gä—9‘9˜e WÐ-°2Ô6€LØÐr#   c           	      óŽ  — t        j                  | |j                  dddd«      «      }|j                  d   }|t	        j
                  |«      z  }|�6|j                  d«      |j                  d«      }
}	|||
|	z
  |
…d |
…f   dz  z  }|�||z   }t        j                  |d¬«      }|�||z  }t        j                  ||«      }||fS )	Nr   r   r
   r   r'   éþÿÿÿg     ˆÃÀr(   )r   ÚmatmulÚpermuteÚshapeÚnpÚsqrtÚsizeÚsoftmax)ÚqÚkÚvÚmaskÚattention_maskÚ	head_maskÚ	matmul_qkÚdkÚscaled_attention_logitsÚndÚnsÚattention_weightsÚoutputs                r!   Úscaled_dot_product_attentionrM   :   sá   € ä—‘˜Q §	¡	¨!¨Q°°1Ó 5Ó6€Ià	
�‰�‰€BØ'¬"¯'©'°"«+Ñ5ÐàÐØ(×-Ñ-¨bÓ1Ð3J×3OÑ3OÐPRÓ3SˆBˆØ 4¨¨R©°"¨°c°r°cÐ(9Ñ#:¸TÑ#AÑAÐàÐ!à"9¸NÑ"JÐäŸ™Ð&=À2ÔFÐð ÐØ-°	Ñ9Ðä�\‰\Ð+¨QÓ/€FàÐ$Ð$Ð$r#   c                   ó<   ‡ — e Zd Zˆ fd„Zd„ Zd„ Z	 	 	 	 	 dd„Zˆ xZS )ÚMultiHeadAttentionc                 ón  •— t         ‰| �  «        || _        || _        t	        || j                  z  «      | _        t        j                  ||«      | _        t        j                  ||«      | _	        t        j                  ||«      | _
        t        j                  ||«      | _        t        «       | _        y ©N)ÚsuperÚ__init__Ú	num_headsr   ÚintÚdepthr   ÚLinearÚWqÚWkÚWvÚdenseÚsetÚpruned_heads)Úselfr   rT   Ú	__class__s      €r!   rS   zMultiHeadAttention.__init__U   s„   ø€ Ü‰ÑÔØ"ˆŒØ(ˆÔä˜¨¯©Ñ6Ó7ˆŒ
ä—)‘)˜L¨,Ó7ˆŒÜ—)‘)˜L¨,Ó7ˆŒÜ—)‘)˜L¨,Ó7ˆŒä—Y‘Y˜|¨\Ó:ˆŒ
Ü›EˆÕr#   c                 ó  — | j                   | j                  z  }t        |«      dk(  ry t        || j                  || j                  «      \  }}t        | j                  |«      | _        t        | j                  |«      | _        t        | j                  |«      | _        t        | j                  |d¬«      | _	        | j                  t        |«      z
  | _        || j                  z  | _         | j                  j                  |«      | _        y )Nr   r   r(   )r   rT   Úlenr   r]   r   rX   rY   rZ   r[   Úunion)r^   ÚheadsÚattention_head_sizeÚindexs       r!   Úprune_headszMultiHeadAttention.prune_headsc   sÑ   € Ø"×/Ñ/°4·>±>ÑAÐÜˆu‹:˜Š?ØÜ7¸¸t¿~¹~ÐObÐdh×duÑduÓv‰ˆˆuô % T§W¡W¨eÓ4ˆŒÜ$ T§W¡W¨eÓ4ˆŒÜ$ T§W¡W¨eÓ4ˆŒÜ'¨¯
©
°E¸qÔAˆŒ
ð Ÿ™¬#¨e«*Ñ4ˆŒØ/°$·.±.Ñ@ˆÔØ ×-Ñ-×3Ñ3°EÓ:ˆÕr#   c                 óx   — |j                  |d| j                  | j                  «      }|j                  g d¢«      S )Nr'   ©r   r   r   r
   )ÚreshaperT   rV   r:   )r^   ÚxÚ
batch_sizes      r!   Úsplit_into_headsz#MultiHeadAttention.split_into_headst   s-   € Ø�I‰I�j " d§n¡n°d·j±jÓAˆØ�y‰yšÓ&Ð&r#   c
                 óx  — |j                   d   }
| j                  |«      }| j                  |«      }| j                  |«      }| j	                  ||
«      }| j	                  ||
«      }| j	                  ||
«      }|�<|d   |d   }}t        j                  ||fd¬«      }t        j                  ||fd¬«      }|du rt        j                  ||f«      }nd}t        ||||||«      }|d   j                  g d¢«      }|d   }|j                  |
d| j                  «      }| j                  |«      }||f}|	r||fz   }|S )	Nr   r   r8   r(   TrQ   rh   r'   )r;   rX   rY   rZ   rl   r   r0   ÚstackrM   r:   ri   r   r[   )r^   rB   rA   r@   rC   Ú
layer_pastrD   rE   Ú	use_cacheÚoutput_attentionsrk   Úpast_keyÚ
past_valueÚpresentrL   Úscaled_attentionÚattnÚoriginal_size_attentionÚoutputss                      r!   ÚforwardzMultiHeadAttention.forwardx   sF  € ð —W‘W˜Q‘Zˆ
à�G‰G�A‹JˆØ�G‰G�A‹JˆØ�G‰G�A‹Jˆà×!Ñ! ! ZÓ0ˆØ×!Ñ! ! ZÓ0ˆØ×!Ñ! ! ZÓ0ˆØÐ!Ø#-¨a¡=°*¸Q±-�jˆHÜ—	‘	˜8 Q˜-¨RÔ0ˆAÜ—	‘	˜: q˜/¨rÔ2ˆAà˜ÑÜ—k‘k 1 a &Ó)‰GàˆGä-¨a°°A°t¸^ÈYÓWˆØ! !™9×,Ñ,ª\Ó:ÐØ�a‰yˆØ"2×":Ñ":¸:ÀrÈ4×K\ÑK\Ó"]ÐØ—‘Ð3Ó4ˆà˜7Ð#ˆÙØ  Ñ'ˆGØˆr#   ©NNNFF)Ú__name__Ú
__module__Ú__qualname__rS   rf   rl   ry   Ú__classcell__©r_   s   @r!   rO   rO   T   s(   ø„ ô"ò;ò"'ð ØØØØ÷(r#   rO   c                 ó¤   — t        j                  t        j                  | |«      t        j                  «       t        j                  || «      «      S rQ   )r   Ú
SequentialrW   ÚReLU)r   Údffs     r!   Úpoint_wise_feed_forward_networkr„   £   s2   € Ü�=‰=œŸ™ <°Ó5´r·w±w³yÄ"Ç)Á)ÈCÐQ]ÓB^Ó_Ð_r#   c                   ó*   ‡ — e Zd Zdˆ fd„	Z	 dd„Zˆ xZS )ÚEncoderLayerc                 ó>  •— t         ‰| �  «        t        ||«      | _        t	        ||«      | _        t        j                  |d¬«      | _        t        j                  |d¬«      | _	        t        j                  |«      | _        t        j                  |«      | _        y )Ng�íµ ÷Æ°>©Úeps)rR   rS   rO   Úmulti_head_attentionr„   Úffnr   Ú	LayerNormÚ
layernorm1Ú
layernorm2ÚDropoutÚdropout1Údropout2)r^   r   rT   rƒ   Úrater_   s        €r!   rS   zEncoderLayer.__init__¨   so   ø€ Ü‰ÑÔä$6°|ÀYÓ$OˆÔ!Ü2°<ÀÓEˆŒäŸ,™, |¸Ô>ˆŒÜŸ,™, |¸Ô>ˆŒäŸ
™
 4Ó(ˆŒÜŸ
™
 4Ó(ˆ�r#   c                 ó  — | j                  |«      }| j                  |||||||||¬«	      }	|	d   }
| j                  |
«      }
||
z   }| j                  |«      }| j	                  |«      }| j                  |«      }||z   }|f|	dd  z   }|S )N©ro   rD   rE   rp   rq   r   r   )r�   rŠ   r�   rŽ   r‹   r‘   )r^   rj   rC   ro   rD   rE   rp   rq   ÚnormedÚattn_outputsÚattn_outputÚout1Úout2Ú
ffn_outputrx   s                  r!   ry   zEncoderLayer.forward´   s¯   € ð —‘ Ó#ˆØ×0Ñ0ØØØØØ!Ø)ØØØ/ð 1ó 

ˆð # 1‘oˆØ—m‘m KÓ0ˆØ�;‰ˆà�‰˜tÓ$ˆØ—X‘X˜d“^ˆ
Ø—]‘] :Ó.ˆ
Ø�jÑ ˆà�'˜L¨¨Ð,Ñ,ˆØˆr#   )gš™™™™™¹?rz   )r{   r|   r}   rS   ry   r~   r   s   @r!   r†   r†   §   s   ø„ õ
)ð qv÷r#   r†   c                   ó   — e Zd ZdZeZdZd„ Zy)ÚCTRLPreTrainedModelz†
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    Útransformerc                 ó  — t        |t        j                  t        f«      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                  «      rJ|j                  j
                  j                  «        |j                  j
                  j                  d«       yy)zInitialize the weights.g        )ÚmeanÚstdNç      ð?)Ú
isinstancer   rW   r   ÚweightÚdataÚnormal_ÚconfigÚinitializer_rangeÚbiasÚzero_Ú	EmbeddingÚpadding_idxrŒ   Úfill_)r^   Úmodules     r!   Ú_init_weightsz!CTRLPreTrainedModel._init_weightsÙ   s  € ä�fœrŸy™y¬&Ð1Ô2ð �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-Ø�K‰K×Ñ×"Ñ"Ô$Ø�M‰M×Ñ×$Ñ$ SÕ)ð .r#   N)r{   r|   r}   Ú__doc__r   Úconfig_classÚbase_model_prefixr®   © r#   r!   rœ   rœ   Ð   s   „ ñð
 €LØ%Ðó*r#   rœ   a>  

    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 ([`CTRLConfig`]): Model configuration class with all the parameters of the model.
            Initializing with a config file does not load the weights associated with the model, only the
            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
a[  
    Args:
        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0].shape[-2]`
            (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary.

            If `past_key_values` is used, only input IDs that do not have their past calculated should be passed as
            `input_ids`.

            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.__call__`] and
            [`PreTrainedTokenizer.encode`] for details.

            [What are input IDs?](../glossary#input-ids)
        past_key_values (`Tuple[Tuple[torch.FloatTensor]]` of length `config.n_layers`):
            Contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see
            `past_key_values` output below). Can be used to speed up sequential decoding. The `input_ids` which have
            their past given to this model should not be passed as input ids as they have already been computed.
        attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.

            [What are attention masks?](../glossary#attention-mask)
        token_type_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
            1]`:

            - 0 corresponds to a *sentence A* token,
            - 1 corresponds to a *sentence B* token.

            [What are token type IDs?](../glossary#token-type-ids)
        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
            config.max_position_embeddings - 1]`.

            [What are position IDs?](../glossary#position-ids)
        head_mask (`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 `(batch_size, sequence_length, hidden_size)`, *optional*):
            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
            model's internal embedding lookup matrix.
        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`).
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        return_dict (`bool`, *optional*):
            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
z^The bare CTRL Model transformer outputting raw hidden-states without any specific head on top.c                   ó®  ‡ — e Zd Zˆ fd„Zd„ Zd„ Zd„ Z ee«       e	e
e¬«      	 	 	 	 	 	 	 	 	 	 	 ddeej                     dee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   deeej&                     e
f   fd„«       «       Zˆ xZS )Ú	CTRLModelc                 óÞ  •— t         ‰| �  |«       |j                  | _        |j                  | _        t        |j                  | j                  t        j                  «      | _
        t        j                  |j                  |j                  «      | _        t        j                  |j                   «      | _        t        j$                  t'        |j                  «      D �cg c]8  }t)        |j                  |j*                  |j,                  |j.                  «      ‘Œ: c}«      | _        t        j2                  |j                  |j4                  ¬«      | _        | j9                  «        y c c}w )Nrˆ   )rR   rS   Ún_embdr   Ún_layerÚ
num_layersr6   Ún_positionsr   Úfloatr5   r   rª   Ú
vocab_sizeÚwr�   Ú
embd_pdropÚdropoutÚ
ModuleListÚranger†   Ún_headrƒ   Úresid_pdropÚhrŒ   Úlayer_norm_epsilonÚ	layernormÚ	post_init)r^   r¦   Ú_r_   s      €r!   rS   zCTRLModel.__init__<  sõ   ø€ Ü‰Ñ˜Ô à"ŸM™MˆÔØ Ÿ.™.ˆŒä/°×0BÑ0BÀD×DUÑDUÔW\×WbÑWbÓcˆÔä—‘˜f×/Ñ/°·±Ó?ˆŒä—z‘z &×"3Ñ"3Ó4ˆŒÜ—‘ÜafÐgm×guÑguÓavÖwÐ\]Œ\˜&Ÿ-™-¨¯©¸¿
¹
ÀF×DVÑDVÕWÒwó
ˆŒô Ÿ™ f§m¡m¸×9RÑ9RÔSˆŒð 	�‰Õùò xs   Ã =E*c                 ó   — | j                   S rQ   ©r¼   ©r^   s    r!   Úget_input_embeddingszCTRLModel.get_input_embeddingsO  s   € Ø�v‰vˆr#   c                 ó   — || _         y rQ   rÉ   ©r^   Únew_embeddingss     r!   Úset_input_embeddingszCTRLModel.set_input_embeddingsR  s	   € Øˆ�r#   c                 ó„   — |j                  «       D ]-  \  }}| j                  |   j                  j                  |«       Œ/ y)zv
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
        N)ÚitemsrÃ   rŠ   rf   )r^   Úheads_to_pruneÚlayerrc   s       r!   Ú_prune_headszCTRLModel._prune_headsU  s>   € ð +×0Ñ0Ó2ò 	B‰LˆE�5Ø�F‰F�5‰M×.Ñ.×:Ñ:¸5ÕAñ	Br#   ©Úoutput_typer°   Ú	input_idsÚpast_key_valuesrD   Útoken_type_idsÚposition_idsrE   Úinputs_embedsrp   rq   Úoutput_hidden_statesÚreturn_dictÚreturnc           
      ó  — |	�|	n| j                   j                  }	|�|n| j                   j                  }|
�|
n| j                   j                  }
|�|n| j                   j                  }|�|�t        d«      ‚|�G| j                  ||«       |j                  «       }|j                  d|d   «      }|j                  d   }n0|�#|j                  «       dd }|j                  d   }nt        d«      ‚|�|j                  n|j                  }|€%d}t        dgt        | j                  «      z  «      }n|d   d   j                  d«      }|€>t        j                  ||d   |z   t        j                   |¬«      }|j#                  d«      }|��|dk  rt        d«      ‚|j                  |d«      }|j#                  d	«      j#                  d
«      }|j%                  | j&                  ¬«      }d|z
  t        j(                  | j&                  «      j*                  z  }| j-                  || j                   j.                  «      }|�I|j                  d|d   «      }| j1                  |«      }|t3        j4                  | j6                  «      z  }nd}|€| j1                  |«      }|d   }t        j8                  t        j:                  ||z   ||z   «      d	«      j%                  |«      }|t3        j4                  | j6                  «      z  }| j<                  j%                  |«      | _        | j<                  |dd…f   }||z   |z   }| j?                  |«      }|rdnd}|
rdnd}|	rdnd}tA        tC        | j                  |«      «      D ]@  \  }\  }}|
r||fz   } |||||||   ||	¬«      }|dd
 \  }}|du r||fz   }|	sŒ8||d
   fz  }ŒB | jE                  |«      }|
r||fz   }|st        d„ ||||fD «       «      S tG        ||||¬«      S )a³  
        Returns:

        Example:

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

        >>> tokenizer = AutoTokenizer.from_pretrained("Salesforce/ctrl")
        >>> model = CTRLModel.from_pretrained("Salesforce/ctrl")

        >>> # CTRL was trained with control codes as the first token
        >>> inputs = tokenizer("Opinion My dog is cute", return_tensors="pt")
        >>> assert inputs["input_ids"][0, 0].item() in tokenizer.control_codes.values()

        >>> outputs = model(**inputs)

        >>> last_hidden_states = outputs.last_hidden_state
        >>> list(last_hidden_states.shape)
        [1, 5, 1280]
        ```NzDYou cannot specify both input_ids and inputs_embeds at the same timer'   r   z5You have to specify either input_ids or inputs_embedsr8   )r&   Údevicez$batch_size has to be defined and > 0r   r   r%   r¡   r²   r”   Tc              3   ó&   K  — | ]	  }|€Œ|–— Œ y ­wrQ   r²   )Ú.0rB   s     r!   ú	<genexpr>z$CTRLModel.forward.<locals>.<genexpr>ë  s   è ø€ Òr˜qÐdeÑdqœÑrùs   ‚Š)Úlast_hidden_staterØ   Úhidden_statesÚ
attentions)$r¦   rq   rp   rÜ   Úuse_return_dictÚ
ValueErrorÚ%warn_if_padding_and_no_attention_maskr>   Úviewr;   rà   Útuplera   rÃ   r   r*   Úlongr-   r,   r&   ÚfinfoÚminÚget_head_maskr·   r¼   r<   r=   r   ÚtriuÚonesr5   r¾   Ú	enumerateÚziprÅ   r   )r^   r×   rØ   rD   rÙ   rÚ   rE   rÛ   rp   rq   rÜ   rÝ   ÚkwargsÚinput_shaperk   rà   Úpast_lengthÚtoken_type_embedsÚseq_lenrC   Ú
pos_embedsrå   ÚpresentsÚall_hidden_statesÚall_attentionsr   rÃ   ro   rx   rt   s                                 r!   ry   zCTRLModel.forward\  s9  € ðN 2CÐ1NÑ-ÐTX×T_ÑT_×TqÑTqÐØ!*Ð!6‘I¸D¿K¹K×<QÑ<Qˆ	à$8Ð$DÑ È$Ï+É+×JjÑJjð 	ð &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆàÐ  ]Ð%>ÜÐcÓdÐdØÐ"Ø×6Ñ6°yÀ.ÔQØ#Ÿ.™.Ó*ˆKØ!Ÿ™ r¨;°r©?Ó;ˆIØ"Ÿ™¨Ñ+‰JØÐ&Ø'×,Ñ,Ó.¨s°Ð3ˆKØ&×,Ñ,¨QÑ/‰JäÐTÓUÐUà%.Ð%:�×!Ò!À×@TÑ@TˆàÐ"ØˆKÜ# T F¬S°·±«[Ñ$8Ó9‰Oà)¨!Ñ,¨QÑ/×4Ñ4°RÓ8ˆKØÐÜ Ÿ<™<¨°[À±_À{Ñ5RÔZ_×ZdÑZdÐmsÔtˆLØ'×1Ñ1°!Ó4ˆLð Ð%Ø˜QŠÜ Ð!GÓHÐHØ+×0Ñ0°¸RÓ@ˆNð ,×5Ñ5°aÓ8×BÑBÀ1ÓEˆNð ,×.Ñ.°T·Z±ZÐ.Ó@ˆNØ! NÑ2´e·k±kÀ$Ç*Á*Ó6M×6QÑ6QÑQˆNð ×&Ñ& y°$·+±+×2EÑ2EÓFˆ	àÐ%Ø+×0Ñ0°°[À±_ÓEˆNØ $§¡ ~Ó 6ÐØ¤§¡¨×):Ñ):Ó!;Ñ;Ñà !ÐàÐ Ø ŸF™F 9Ó-ˆMà˜b‘/ˆÜ�z‰zœ%Ÿ*™* W¨{Ñ%:¸GÀkÑ<QÓRÐTUÓV×YÑYÐZ`ÓaˆàœŸ™ ×!2Ñ!2Ó3Ñ3ˆð !×-Ñ-×0Ñ0°Ó8ˆÔØ×&Ñ& |²Q Ñ7ˆ
à%¨
Ñ2Ð5FÑFˆàŸ™ ]Ó3ˆá"‘2¨ˆÙ"6™B¸DÐÙ0™°dˆÜ"+¬C°·±¸Ó,HÓ"Iò 	0ÑˆA‰��:Ù#Ø$5¸Ð8HÑ$HÐ!ÙØØØ%Ø-Ø# A™,Ø#Ø"3ôˆGð &-¨R¨a [Ñ"ˆM˜7Ø˜DÑ Ø# w jÑ0�â Ø 7¨1¡: -Ñ/‘ð#	0ð& Ÿ™ }Ó5ˆÙØ 1°]Ð4DÑ DÐáÜÑr ]°HÐ>OÐQ_Ð$`ÔrÓrÐrä&Ø+Ø$Ø+Ø%ô	
ð 	
r#   )NNNNNNNNNNN)r{   r|   r}   rS   rË   rÏ   rÔ   r   ÚCTRL_INPUTS_DOCSTRINGr   r   Ú_CONFIG_FOR_DOCr   r   Ú
LongTensorr   ÚFloatTensorÚboolr   ÚTensorry   r~   r   s   @r!   r´   r´   7  sg  ø„ ô
ò&ò òBñ +Ð+@ÓAÙÐ+BÐQ`Ôað 15ØEIØ6:Ø59Ø37Ø15Ø59Ø$(Ø,0Ø/3Ø&*ñT
à˜E×,Ñ,Ñ-ðT
ð " %¨¨e×.?Ñ.?Ñ(@Ñ"AÑBðT
ð ! ×!2Ñ!2Ñ3ð	T
ð
 ! ×!1Ñ!1Ñ2ðT
ð ˜u×/Ñ/Ñ0ðT
ð ˜E×-Ñ-Ñ.ðT
ð   × 1Ñ 1Ñ2ðT
ð ˜D‘>ðT
ð $ D™>ðT
ð ' t™nðT
ð ˜d‘^ðT
ð 
ˆu�U—\‘\Ñ"Ð$;Ð;Ñ	<òT
ó bó BôT
r#   r´   z‡
    The CTRL Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    c                   óH  ‡ — e Zd ZdgZˆ fd„Zd„ Zd„ Z ee«       e	e
e¬«      	 	 	 	 	 	 	 	 	 	 	 	 ddeej                     deeeej                            deej                      d	eej                     d
eej                     deej                      deej                      deej                     dee   dee   dee   dee   deeej&                     e
f   fd„«       «       Zdd„Zedeeej&                        dej&                  deeej&                        fd„«       Zˆ xZS )ÚCTRLLMHeadModelzlm_head.weightc                 óÆ   •— t         ‰| �  |«       t        |«      | _        t	        j
                  |j                  |j                  d¬«      | _        | j                  «        y )NT©r¨   )
rR   rS   r´   r�   r   rW   r¶   r»   Úlm_headrÆ   ©r^   r¦   r_   s     €r!   rS   zCTRLLMHeadModel.__init__ÿ  sG   ø€ Ü‰Ñ˜Ô Ü$ VÓ,ˆÔÜ—y‘y §¡°×0AÑ0AÈÔMˆŒð 	�‰Õr#   c                 ó   — | j                   S rQ   ©r  rÊ   s    r!   Úget_output_embeddingsz%CTRLLMHeadModel.get_output_embeddings  s   € Ø�|‰|Ðr#   c                 ó   — || _         y rQ   r
  rÍ   s     r!   Úset_output_embeddingsz%CTRLLMHeadModel.set_output_embeddings
  s	   € Ø%ˆ�r#   rÕ   r×   rØ   rD   rÙ   rÚ   rE   rÛ   Úlabelsrp   rq   rÜ   rÝ   rÞ   c                 ó|  — |�|n| j                   j                  }| j                  ||||||||	|
||¬«      }|d   }| j                  |«      }d}|�* | j                  ||fd| j                   j
                  i|¤Ž}|s|f|dd z   }|�|f|z   S |S t        |||j                  |j                  |j                  ¬«      S )ay  
        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]`

        Returns:

        Example:

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

        >>> tokenizer = AutoTokenizer.from_pretrained("Salesforce/ctrl")
        >>> model = CTRLLMHeadModel.from_pretrained("Salesforce/ctrl")

        >>> # CTRL was trained with control codes as the first token
        >>> inputs = tokenizer("Wikipedia The llama is", return_tensors="pt")
        >>> assert inputs["input_ids"][0, 0].item() in tokenizer.control_codes.values()

        >>> sequence_ids = model.generate(inputs["input_ids"])
        >>> sequences = tokenizer.batch_decode(sequence_ids)
        >>> sequences
        ['Wikipedia The llama is a member of the family Bovidae. It is native to the Andes of Peru,']

        >>> outputs = model(**inputs, labels=inputs["input_ids"])
        >>> round(outputs.loss.item(), 2)
        9.21

        >>> list(outputs.logits.shape)
        [1, 5, 246534]
        ```N©
rØ   rD   rÙ   rÚ   rE   rÛ   rp   rq   rÜ   rÝ   r   r»   r   )ÚlossÚlogitsrØ   rå   ræ   )
r¦   rç   r�   r  Úloss_functionr»   r   rØ   rå   ræ   )r^   r×   rØ   rD   rÙ   rÚ   rE   rÛ   r  rp   rq   rÜ   rÝ   rô   Útransformer_outputsrå   Ú	lm_logitsr  rL   s                      r!   ry   zCTRLLMHeadModel.forward  s  € ðf &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà"×.Ñ.ØØ+Ø)Ø)Ø%ØØ'ØØ/Ø!5Ø#ð /ó 
Ðð ,¨AÑ.ˆà—L‘L Ó/ˆ	àˆØÐØ%�4×%Ñ%ØØñð  Ÿ;™;×1Ñ1ðð ñ	ˆDñ Ø�\Ð$7¸¸Ð$;Ñ;ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä%ØØØ/×?Ñ?Ø-×;Ñ;Ø*×5Ñ5ô
ð 	
r#   c                 ó    — |�G|d   d   j                   d   }|j                   d   |kD  r|}n|j                   d   dz
  }|d d …|d …f   }|||dœS )Nr   r   r   )r×   rØ   rp   )r;   )r^   r×   rØ   rp   rô   rö   Úremove_prefix_lengths          r!   Úprepare_inputs_for_generationz-CTRLLMHeadModel.prepare_inputs_for_generationi  st   € ð Ð&Ø)¨!Ñ,¨QÑ/×5Ñ5°aÑ8ˆKð �‰˜qÑ! KÒ/Ø'2Ñ$ð (1§¡°qÑ'9¸AÑ'=Ð$à!¢!Ð%9Ñ%:Ð":Ñ;ˆIà&¸?ÐYbÑcÐcr#   Úbeam_idxc                 ó,   ‡— t        ˆfd„| D «       «      S )a  
        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
        beam_idx at every generation step.
        c              3   óF   •K  — | ]  }t        ˆfd „|D «       «      –— Œ y­w)c              3   ót   •K  — | ]/  }|j                  d ‰j                  |j                  «      «      –— Œ1 y­w)r   N)Úindex_selectr,   rà   )râ   Ú
past_stater  s     €r!   rã   z;CTRLLMHeadModel._reorder_cache.<locals>.<genexpr>.<genexpr>…  s.   øè ø€ ÒjÐQ[�*×)Ñ)¨!¨X¯[©[¸×9JÑ9JÓ-K×LÑjùs   ƒ58N©rë   )râ   ro   r  s     €r!   rã   z1CTRLLMHeadModel._reorder_cache.<locals>.<genexpr>„  s%   øè ø€ ò 
àô ÓjÐ_iÔj×jñ
ùs   ƒ!r  )rØ   r  s    `r!   Ú_reorder_cachezCTRLLMHeadModel._reorder_cache{  s   ø€ ô ó 
à-ô
ó 
ð 	
r#   ©NNNNNNNNNNNN©NN)r{   r|   r}   Ú_tied_weights_keysrS   r  r  r   rý   r   r   rþ   r   r   rÿ   r   r   r  r   r  ry   r  Ústaticmethodr   r~   r   s   @r!   r  r  õ  sØ  ø„ ð +Ð+Ðôòò&ñ +Ð+@ÓAÙÐ+AÐP_Ô`ð 15ØEIØ6:Ø59Ø37Ø15Ø59Ø-1Ø$(Ø,0Ø/3Ø&*ñX
à˜E×,Ñ,Ñ-ðX
ð " %¨¨e×.?Ñ.?Ñ(@Ñ"AÑBðX
ð ! ×!2Ñ!2Ñ3ð	X
ð
 ! ×!1Ñ!1Ñ2ðX
ð ˜u×/Ñ/Ñ0ðX
ð ˜E×-Ñ-Ñ.ðX
ð   × 1Ñ 1Ñ2ðX
ð ˜×)Ñ)Ñ*ðX
ð ˜D‘>ðX
ð $ D™>ðX
ð ' t™nðX
ð ˜d‘^ðX
ð 
ˆu�U—\‘\Ñ"Ð$:Ð:Ñ	;òX
ó aó BðX
ótdð$ ð
Ø˜u U§\¡\Ñ2Ñ3ð
Ø?D¿|¹|ð
à	ˆu�U—\‘\Ñ"Ñ	#ò
ó ô
r#   r  aÎ  
    The CTRL Model transformer with a sequence classification head on top (linear layer).
    [`CTRLForSequenceClassification`] 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                   ó¼  ‡ — e Zd Zˆ fd„Z ee«       eee¬«      	 	 	 	 	 	 	 	 	 	 	 	 dde	e
j                     de	e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   deee
j                      ef   fd„«       «       Zˆ xZS )ÚCTRLForSequenceClassificationc                 óè   •— t         ‰| �  |«       |j                  | _        t        |«      | _        t        j                  |j                  | j                  d¬«      | _        | j                  «        y )NFr  )
rR   rS   Ú
num_labelsr´   r�   r   rW   r¶   Ú
classifierrÆ   r  s     €r!   rS   z&CTRLForSequenceClassification.__init__—  sR   ø€ Ü‰Ñ˜Ô Ø ×+Ñ+ˆŒÜ$ VÓ,ˆÔÜŸ)™) F§M¡M°4·?±?ÈÔOˆŒð 	�‰Õr#   rÕ   r×   rØ   rD   rÙ   rÚ   rE   rÛ   r  rp   rq   rÜ   rÝ   rÞ   c                 óÀ  — |�|n| j                   j                  }| j                  ||||||||	|
||¬«      }|d   }| j                  |«      }|�|j                  dd \  }}n|j                  dd \  }}| j                   j
                  €|dk7  rt        d«      ‚| j                   j
                  €d}nÃ|�“|| j                   j
                  k7  j                  |j                  t        j                  «      }t        j                  |j                  d   |j                  t        j                  ¬«      }||z  j                  d«      }n.d}t        j                  | j                  j                   › d	�«       |t        j                  ||j                  ¬
«      |f   }d}|��‡| 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$                  dk(  r& ||j/                  «       |j/                  «       «      }nŒ |||«      }n‚| j                   j"                  dk(  r=t1        «       } ||j3                  d| j$                  «      |j3                  d«      «      }n,| j                   j"                  dk(  rt5        «       } |||«      }|s|f|dd z   }|�|f|z   S |S t7        |||j8                  |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).

        Returns:

        Example of single-label classification:

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

        >>> tokenizer = AutoTokenizer.from_pretrained("Salesforce/ctrl")
        >>> model = CTRLForSequenceClassification.from_pretrained("Salesforce/ctrl")

        >>> # CTRL was trained with control codes as the first token
        >>> inputs = tokenizer("Opinion My dog is cute", return_tensors="pt")
        >>> assert inputs["input_ids"][0, 0].item() in tokenizer.control_codes.values()

        >>> with torch.no_grad():
        ...     logits = model(**inputs).logits

        >>> predicted_class_id = logits.argmax().item()
        >>> model.config.id2label[predicted_class_id]
        'LABEL_0'
        ```

        ```python
        >>> import torch

        >>> torch.manual_seed(42)  # doctest: +IGNORE_RESULT
        >>> # To train a model on `num_labels` classes, you can pass `num_labels=num_labels` to `.from_pretrained(...)`
        >>> num_labels = len(model.config.id2label)
        >>> model = CTRLForSequenceClassification.from_pretrained("Salesforce/ctrl", num_labels=num_labels)

        >>> labels = torch.tensor(1)
        >>> loss = model(**inputs, labels=labels).loss
        >>> round(loss.item(), 2)
        0.93
        ```

        Example of multi-label classification:

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

        >>> tokenizer = AutoTokenizer.from_pretrained("Salesforce/ctrl")
        >>> model = CTRLForSequenceClassification.from_pretrained(
        ...     "Salesforce/ctrl", problem_type="multi_label_classification"
        ... )

        >>> # CTRL was trained with control codes as the first token
        >>> inputs = tokenizer("Opinion My dog is cute", return_tensors="pt")
        >>> assert inputs["input_ids"][0, 0].item() in tokenizer.control_codes.values()

        >>> with torch.no_grad():
        ...     logits = model(**inputs).logits

        >>> predicted_class_id = logits.argmax().item()
        >>> model.config.id2label[predicted_class_id]
        'LABEL_0'
        ```

        ```python
        >>> # To train a model on `num_labels` classes, you can pass `num_labels=num_labels` to `.from_pretrained(...)`
        >>> num_labels = len(model.config.id2label)
        >>> model = CTRLForSequenceClassification.from_pretrained("Salesforce/ctrl", num_labels=num_labels)

        >>> num_labels = len(model.config.id2label)
        >>> labels = torch.nn.functional.one_hot(torch.tensor([predicted_class_id]), num_classes=num_labels).to(
        ...     torch.float
        ... )
        >>> loss = model(**inputs, labels=labels).loss
        >>> loss.backward()  # doctest: +IGNORE_RESULT
        ```Nr  r   r   r   z=Cannot handle batch sizes > 1 if no padding token is defined.r'   )rà   r&   zŠ will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`)rà   Ú
regressionÚsingle_label_classificationÚmulti_label_classification)r  r  rå   ræ   )r¦   rç   r�   r)  r;   Úpad_token_idrè   r,   rà   r   Úint32r*   ÚargmaxÚloggerÚwarning_oncer_   r{   Úproblem_typer(  r&   rì   rU   r	   Úsqueezer   rê   r   r   rå   ræ   )r^   r×   rØ   rD   rÙ   rÚ   rE   rÛ   r  rp   rq   rÜ   rÝ   r  rå   r  rk   Úsequence_lengthÚlast_non_pad_tokenÚnon_pad_maskÚtoken_indicesÚpooled_logitsr  Úloss_fctrL   s                            r!   ry   z%CTRLForSequenceClassification.forward   s  € ð@ &1Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà"×.Ñ.ØØ+Ø)Ø)Ø%ØØ'ØØ/Ø!5Ø#ð /ó 
Ðð ,¨AÑ.ˆØ—‘ Ó/ˆàÐ Ø*3¯/©/¸"¸1Ð*=Ñ'ˆJ™à*7×*=Ñ*=¸b¸qÐ*AÑ'ˆJ˜à�;‰;×#Ñ#Ð+°
¸a²ÜÐ\Ó]Ð]Ø�;‰;×#Ñ#Ð+Ø!#ÑØÐ"à%¨¯©×)AÑ)AÑA×EÑEÀfÇmÁmÔUZ×U`ÑU`ÓaˆLÜ!ŸL™L¨¯©¸Ñ)<ÀVÇ]Á]ÔZ_×ZeÑZeÔfˆMØ"/°,Ñ">×!FÑ!FÀrÓ!JÑà!#ÐÜ×ÑØ—>‘>×*Ñ*Ð+ð ,Zð Zôð
 œuŸ|™|¨J¸v¿}¹}ÔMÐOaÐaÑbˆàˆØÑØ�{‰{×'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# M×$9Ñ$9Ó$;¸V¿^¹^Ó=MÓN‘Dá# M°6Ó:‘DØ—‘×)Ñ)Ð-JÒJÜ+Ó-�Ù × 2Ñ 2°2°t·±Ó GÈÏÉÐUWËÓY‘Ø—‘×)Ñ)Ð-IÒIÜ,Ó.�Ù ¨vÓ6�ÙØ#Ð%Ð(;¸A¸BÐ(?Ñ?ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä'ØØ Ø-×;Ñ;Ø*×5Ñ5ô	
ð 	
r#   r!  )r{   r|   r}   rS   r   rý   r   r   rþ   r   r   rÿ   r   r   r  r   r  ry   r~   r   s   @r!   r&  r&  Š  sp  ø„ ôñ +Ð+@ÓAÙÐ+CÐRaÔbð 15ØEIØ6:Ø59Ø37Ø15Ø59Ø-1Ø$(Ø,0Ø/3Ø&*ñg
à˜E×,Ñ,Ñ-ðg
ð " %¨¨e×.?Ñ.?Ñ(@Ñ"AÑBðg
ð ! ×!2Ñ!2Ñ3ð	g
ð
 ! ×!1Ñ!1Ñ2ðg
ð ˜u×/Ñ/Ñ0ðg
ð ˜E×-Ñ-Ñ.ðg
ð   × 1Ñ 1Ñ2ðg
ð ˜×)Ñ)Ñ*ðg
ð ˜D‘>ðg
ð $ D™>ðg
ð ' t™nðg
ð ˜d‘^ðg
ð 
ˆu�U—\‘\Ñ"Ð$<Ð<Ñ	=òg
ó có Bôg
r#   r&  )r&  r  r´   rœ   r"  )2r¯   Útypingr   r   r   Únumpyr<   r   r   Útorch.nnr   r   r	   Ú
generationr   Úmodeling_outputsr   r   r   Úmodeling_utilsr   Úpytorch_utilsr   r   r   Úutilsr   r   r   r   Úconfiguration_ctrlr   Ú
get_loggerr{   r1  rþ   r"   r6   rM   ÚModulerO   r„   r†   rœ   ÚCTRL_START_DOCSTRINGrý   r´   r  r&  Ú__all__r²   r#   r!   ú<module>rH     s<  ðñ  ç )Ñ )ã Û Ý ß AÑ Aå )ß iÑ iÝ -ß YÑ Yß tÓ tÝ *ð 
ˆ×	Ñ	˜HÓ	%€à€òò
ó%ô4L˜Ÿ™ô Lò^`ô&�2—9‘9ô &ôR*˜/ô *ð4Ð ð :Ð ñz ØdØóôw
Ð#ó w
ó	ðw
ñt ðð óôK
Ð)¨?ó K
óðK
ñ\ ðð óôs
Ð$7ó s
óðs
òl c�r#   