Ë
    T^(hôË  ã                   ót  — d Z ddlZddl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 ddlmZmZmZ ddl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$Z$m%Z% ddl&m'Z'  e#jP                  e)«      Z*dZ+dZ,d„ Z- G d„ dej\                  «      Z/ G d„ dej\                  «      Z0 G d„ dej\                  «      Z1 G d„ dej\                  «      Z2 G d„ de«      Z3dZ4dZ5 e!de4«       G d „ d!e3«      «       Z6 e!d"e4«       G d#„ d$e3e«      «       Z7 e!d%e4«       G d&„ d'e3«      «       Z8g d(¢Z9y))zPyTorch OpenAI ImageGPT model.é    N)ÚAnyÚOptionalÚTupleÚUnion)Únn)Úautocast)ÚBCEWithLogitsLossÚCrossEntropyLossÚMSELossé   )ÚACT2FN)ÚGenerationMixin)Ú)BaseModelOutputWithPastAndCrossAttentionsÚ!CausalLMOutputWithCrossAttentionsÚ SequenceClassifierOutputWithPast)ÚPreTrainedModel)ÚConv1DÚ find_pruneable_heads_and_indicesÚprune_conv1d_layer)Úadd_start_docstringsÚ%add_start_docstrings_to_model_forwardÚloggingÚreplace_return_docstringsÚtorch_floaté   )ÚImageGPTConfigzopenai/imagegpt-smallr   c                 óò  — 	 ddl }ddl}t
        j                  j                  |«      }t        j                  dj                  |«      «       |j                  j                  |«      }g }g }|D ]v  \  }	}
t        j                  dj                  |	|
«      «       |j                  j                  ||	«      }|j                  |	«       |j                  |j                  «       «       Œx t        ||«      D �]   \  }	}|	dd }	|	j!                  d«      }	t#        d„ |	D «       «      s|	d	   d
v r4t        j                  dj                  dj%                  |	«      «      «       Œj| }|	d	   dvrt'        |d«      }|	D �]W  }|j)                  d|«      r|j!                  d|«      }n|g}|d   dk(  s|d   dk(  rt'        |d«      }nì|d   dk(  rt'        |d«      }n×|d   dk(  s|d   dk(  rt'        ||d   «      }t'        |d«      }n«|d   dv rt'        |d«      }t'        |d«      }n‹t+        |	«      dk(  r,|	d   dk(  r$|d   dk(  rt'        ||d   «      }t'        |d«      }nQ|d   dk(  rt'        |d«      }t'        |d«      }n0|d   dk(  rt'        |d«      }t'        |d«      }nt'        ||d   «      }t+        |«      d k\  s�ŒEt-        |d   «      }||   }�ŒZ t+        |	«      dkD  r|	d   dk(  s|	d	   dk(  s|	d	   dk(  s|	d	   dk(  rn	 |j.                  |j.                  k(  sJ ‚	 t        j                  d!j                  |	«      «       |	d	   d"k(  rbt5        j6                  |j9                  |j:                  |j:                  «      «      j<                  |j>                  dd…d|j:                  …f<   �Œ¶|	d	   d#k(  rot5        j6                  |j9                  |j:                  |j:                  «      «      j<                  |j>                  dd…|j:                  d |j:                  z  …f<   �Œ-|	d	   d$k(  ret5        j6                  |j9                  |j:                  |j:                  «      «      j<                  |j>                  dd…d |j:                  z  d…f<   �Œšt+        |	«      dk(  rP|	d   dk(  rH|	d    dk(  r@t5        j6                  |j9                  |j:                  |j:                  «      «      |_        �Œø|	d	   dk(  rt5        j6                  |«      |_        �Œ|	d	   dk(  r7t5        j6                  |«      |j>                  d|j@                  dz
  …dd…f<   �Œ[|	d	   dk(  r$t5        j6                  |«      |j>                  d	<   �Œ‡t5        j6                  |«      |_        �Œ£ | S # t        $ r t        j	                  d«       ‚ w xY w# t0        $ r1}|xj2                  |j.                  |j.                  fz  c_        ‚ d}~ww xY w)%z0
    Load tf checkpoints in a pytorch model
    r   Nz™Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see https://www.tensorflow.org/install/ for installation instructions.z(Converting TensorFlow checkpoint from {}z"Loading TF weight {} with shape {}é   ú/c              3   ó$   K  — | ]  }|d v –— Œ
 y­w))Úadam_vÚadam_mÚAdamWeightDecayOptimizerÚAdamWeightDecayOptimizer_1Úglobal_stepN© )Ú.0Úns     úl/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/transformers/models/imagegpt/modeling_imagegpt.pyú	<genexpr>z.load_tf_weights_in_imagegpt.<locals>.<genexpr>V   s   è ø€ ò 
àð ÐnÔnñ
ùó   ‚éÿÿÿÿ)Ú_stepzSkipping {})ÚwtetÚtransformerz[A-Za-z]+\d+z(\d+)ÚwÚgÚweightÚbÚbiasÚwpeÚwte)Úq_projÚk_projÚv_projÚc_attnr   r   ÚattnÚc_projr.   Úlm_headÚsosé   zInitialize PyTorch weight {}r7   r8   r9   )!ÚreÚ
tensorflowÚImportErrorÚloggerÚerrorÚosÚpathÚabspathÚinfoÚformatÚtrainÚlist_variablesÚload_variableÚappendÚsqueezeÚzipÚsplitÚanyÚjoinÚgetattrÚ	fullmatchÚlenÚintÚshapeÚAssertionErrorÚargsÚtorchÚ
from_numpyÚreshapeÚn_embdÚTÚdataÚ
vocab_size)ÚmodelÚconfigÚimagegpt_checkpoint_pathr@   ÚtfÚtf_pathÚ	init_varsÚnamesÚarraysÚnamerW   ÚarrayÚpointerÚm_nameÚscope_namesÚnumÚes                    r)   Úload_tf_weights_in_imagegptrp   5   s�  € ð	Ûãô �g‰g�o‰oÐ6Ó7€GÜ
‡K�KÐ:×AÑAÀ'ÓJÔKà—‘×'Ñ'¨Ó0€IØ€EØ€Fà ò '‰ˆˆeÜ�‰Ð8×?Ñ?ÀÀeÓLÔMØ—‘×&Ñ& w°Ó5ˆØ�‰�TÔØ�‰�e—m‘m“oÕ&ð	'ô ˜5 &Ó)ó L3‰ˆˆeØ�A�BˆxˆØ�z‰z˜#‹ˆô ñ 
àô
ô 
ð �"‰X˜Ñ"Ü�K‰K˜×,Ñ,¨S¯X©X°d«^Ó<Ô=ØàˆØ�‰8˜8Ñ#Ü˜g }Ó5ˆGàó 	'ˆFØ�|‰|˜O¨VÔ4Ø Ÿh™h x°Ó8‘à%˜h�à˜1‰~ Ò$¨°A©¸#Ò(=Ü! '¨8Ó4‘Ø˜Q‘ 3Ò&Ü! '¨6Ó2‘Ø˜Q‘ 5Ò(¨K¸©N¸eÒ,CÜ! '¨;°q©>Ó:�Ü! '¨8Ó4‘Ø˜Q‘Ð#AÑAÜ! '¨8Ó4�Ü! '¨8Ó4‘Ü�T“˜a’ D¨¡G¨vÒ$5¸+Àa¹.ÈHÒ:TÜ! '¨;°q©>Ó:�Ü! '¨8Ó4‘Ø˜Q‘ 6Ò)Ü! '¨9Ó5�Ü! '¨8Ó4‘Ø˜Q‘ 5Ò(Ü! '¨5Ó1�Ü! '¨8Ó4‘ä! '¨;°q©>Ó:�Ü�;Ó 1Ô$Ü˜+ a™.Ó)�Ø! #™,’ð;	'ô> ˆt‹9�qŠ=˜T !™W¨Ò.°$°r±(¸fÒ2DÈÈRÉÐTYÒHYÐ]aÐbdÑ]eÐinÒ]nØðØ—}‘}¨¯©Ò3Ð3Ñ3ô
 	�‰Ð2×9Ñ9¸$Ó?Ô@à�‰8�xÒÜ/4×/?Ñ/?ÀÇÁÈfÏmÉmÐ]c×]jÑ]jÓ@kÓ/l×/nÑ/nˆG�L‰Lš˜O˜fŸm™m˜OÐ+Ó,Ø�"‰X˜Ò!ÜAF×AQÑAQØ—‘˜fŸm™m¨V¯]©]Ó;óBç‰að �L‰Lš˜FŸM™M¨A°·±Ñ,=Ð=Ð=Ó>ð �"‰X˜Ò!Ü38×3CÑ3CÀEÇMÁMÐRX×R_ÑR_Ðag×anÑanÓDoÓ3p×3rÑ3rˆG�L‰Lš˜A §¡Ñ-Ñ/Ð/Ó0Ü�‹Y˜!Š^  Q¡¨6Ò 1°d¸1±gÀÒ6IÜ ×+Ñ+¨E¯M©M¸&¿-¹-ÈÏÉÓ,WÓXˆGŽLØ�"‰X˜ÒÜ ×+Ñ+¨EÓ2ˆGŽLØ�"‰X˜ÒÜ7<×7GÑ7GÈÓ7NˆG�L‰LÐ0˜6×,Ñ,¨qÑ0Ð0²!Ð3Ó4Ø�"‰X˜ÒÜ$×/Ñ/°Ó6ˆG�L‰L˜Óä ×+Ñ+¨EÓ2ˆGŽLðYL3ð\ €LøôC ò Ü�‰ðQô	
ð 	ðûôP "ò Ø—’˜7Ÿ=™=¨%¯+©+Ð6Ñ6•Øûðús#   ‚V Ë?V<Ö V9Ö<	W6×,W1×1W6c                   óh   ‡ — e Zd Zddee   defˆ fd„Zdej                  dej                  fd„Z	ˆ xZ
S )ÚImageGPTLayerNormÚhidden_sizeÚepsc                 óŠ   •— t         ‰| �  «        || _        t        j                  t        j                  |«      «      | _        y ©N)ÚsuperÚ__init__rt   r   Ú	ParameterrZ   ÚTensorr2   )Úselfrs   rt   Ú	__class__s      €r)   rx   zImageGPTLayerNorm.__init__¢   s.   ø€ Ü‰ÑÔØˆŒÜ—l‘l¤5§<¡<°Ó#<Ó=ˆ�ó    ÚtensorÚreturnc                 óÀ   — |t        j                  t        j                  t        j                  |«      dd¬«      | j                  z   «      z  }|| j
                  z  }|S )Nr,   T)ÚaxisÚkeepdim)rZ   ÚsqrtÚmeanÚsquarert   r2   )r{   r~   s     r)   ÚforwardzImageGPTLayerNorm.forward§   sK   € àœ%Ÿ*™*¤U§Z¡Z´·±¸VÓ0DÈ2ÐW[Ô%\Ð_c×_gÑ_gÑ%gÓhÑhˆØ˜$Ÿ+™+Ñ%ˆØˆr}   )gñhãˆµøä>)Ú__name__Ú
__module__Ú__qualname__r   rV   Úfloatrx   rZ   rz   r†   Ú__classcell__©r|   s   @r)   rr   rr   ¡   s5   ø„ ñ> E¨#¡Jð >°Uõ >ð
˜eŸl™lð ¨u¯|©|÷ r}   rr   c                   ó"  ‡ — e Zd Zddee   dee   fˆ fd„Zd„ Zdd„Zdd„Z	d„ Z
d„ Z	 	 	 	 	 	 	 dd	ej                  d
ee   deej                     deej                     deej                     deej                     dee   dee   defd„Zˆ xZS )ÚImageGPTAttentionÚis_cross_attentionÚ	layer_idxc           	      óü  •— t         ‰| �  «        |j                  }| j                  dt	        j
                  t	        j                  ||ft        j                  ¬«      «      j                  dd||«      d¬«       | j                  dt	        j                  d«      d¬«       |j                  | _        |j                  | _        | j                  | j                  z  | _        | j                  | _        | j                  | j                  z  | j                  k7  r&t!        d| j                  › d	| j                  › d
�«      ‚|j"                  | _        || _        |j&                  | _        || _        |j*                  | _        | j$                  rNt-        d| j                  z  | j                  «      | _        t-        | j                  | j                  «      | _        n(t-        d| j                  z  | j                  «      | _        t-        | j                  | j                  «      | _        t5        j6                  |j8                  «      | _        t5        j6                  |j<                  «      | _        tA        «       | _!        y )Nr4   ©Údtyper   F)Ú
persistentÚmasked_biasg     ˆÃÀz=`embed_dim` must be divisible by num_heads (got `embed_dim`: z and `num_heads`: z).r?   r   )"rw   rx   Úmax_position_embeddingsÚregister_bufferrZ   ÚtrilÚonesÚboolÚviewr~   rs   Ú	embed_dimÚnum_attention_headsÚ	num_headsÚhead_dimÚ
split_sizeÚ
ValueErrorÚscale_attn_weightsr�   Úscale_attn_by_inverse_layer_idxr�   Úreorder_and_upcast_attnr   r:   Úq_attnr<   r   ÚDropoutÚ
attn_pdropÚattn_dropoutÚresid_pdropÚresid_dropoutÚsetÚpruned_heads)r{   rb   r�   r�   Úmax_positionsr|   s        €r)   rx   zImageGPTAttention.__init__¯   sí  ø€ Ü‰ÑÔà×6Ñ6ˆØ×ÑØÜ�J‰J”u—z‘z =°-Ð"@ÌÏ
É
ÔSÓT×YÑYØ�1�m ]óð ð 	ô 	
ð 	×Ñ˜]¬E¯L©L¸Ó,>È5ÐÔQà×+Ñ+ˆŒØ×3Ñ3ˆŒØŸ™¨$¯.©.Ñ8ˆŒØŸ.™.ˆŒØ�=‰=˜4Ÿ>™>Ñ)¨T¯^©^Ò;ÜØOÐPT×P^ÑP^ÐO_ð `Ø—N‘NÐ# 2ð'óð ð
 #)×";Ñ";ˆÔØ"4ˆÔð 06×/UÑ/UˆÔ,Ø"ˆŒØ'-×'EÑ'EˆÔ$à×"Ò"Ü   T§^¡^Ñ!3°T·^±^ÓDˆDŒKÜ  §¡°·±Ó@ˆD�Kä   T§^¡^Ñ!3°T·^±^ÓDˆDŒKÜ˜TŸ^™^¨T¯^©^Ó<ˆŒäŸJ™J v×'8Ñ'8Ó9ˆÔÜŸZ™Z¨×(:Ñ(:Ó;ˆÔä›EˆÕr}   c                 óF  — t        |«      dk(  ry t        || j                  | j                  | j                  «      \  }}t        j                  ||| j                  z   |d| j                  z  z   g«      }t        | j                  |d¬«      | _	        t        | j                  |d¬«      | _
        | j                  | j                  z  | j                  t        |«      z
  z  | _        | j                  t        |«      z
  | _        | j                  j                  |«      | _        y )Nr   r?   r   ©Údim)rU   r   rž   rŸ   r¬   rZ   Úcatr    r   r:   r<   Úunion)r{   ÚheadsÚindexÚ
index_attns       r)   Úprune_headszImageGPTAttention.prune_headsÚ   sá   € Üˆu‹:˜Š?ØÜ7¸¸t¿~¹~ÈtÏ}É}Ð^b×^oÑ^oÓp‰ˆˆuÜ—Y‘Y  u¨t¯©Ñ'>ÀÈÈTÏ_É_ÑI\Ñ@]Ð^Ó_ˆ
ô )¨¯©°jÀaÔHˆŒÜ(¨¯©°eÀÔCˆŒð  Ÿ?™?¨d¯n©nÑ<ÀÇÁÔRUÐV[ÓR\ÑA\Ñ]ˆŒØŸ™¬#¨e«*Ñ4ˆŒØ ×-Ñ-×3Ñ3°EÓ:ˆÕr}   c                 óD  — t        j                  ||j                  dd«      «      }| j                  r |t	        |j                  d«      dz  «      z  }| j                  r|t        | j                  dz   «      z  }| j                  s¬|j                  d«      |j                  d«      }}| j                  d d …d d …||z
  |…d |…f   }	t        j                  |j                  «      j                  }
t        j                  |
|j                  |j                  ¬«      }
t        j                   |	||
«      }|�||z   } t#        j$                  d¬«      |«      }|j'                  |j                  «      }| j)                  |«      }|�||z  }t        j                  ||«      }||fS )Nr,   éþÿÿÿç      à?r   ©r“   Údevicer¯   )rZ   ÚmatmulÚ	transposer¢   r   Úsizer£   rŠ   r�   r�   r4   Úfinfor“   Úminr~   r»   Úwherer   ÚSoftmaxÚtyper¨   )r{   ÚqueryÚkeyÚvalueÚattention_maskÚ	head_maskÚattn_weightsÚquery_lengthÚ
key_lengthÚcausal_maskÚ
mask_valueÚattn_outputs               r)   Ú_attnzImageGPTAttention._attné   st  € Ü—|‘| E¨3¯=©=¸¸RÓ+@ÓAˆà×"Ò"Ø'¬+°e·j±jÀ³nÈÑ6KÓ*LÑLˆLð ×/Ò/Ø'¬%°·±ÀÑ0BÓ*CÑCˆLà×&Ò&à',§z¡z°"£~°s·x±xÀ³|˜*ˆLØŸ)™)¢A¢q¨*°|Ñ*CÀjÐ*PÐR]ÐS]ÐR]Ð$]Ñ^ˆKÜŸ™ \×%7Ñ%7Ó8×<Ñ<ˆJô Ÿ™ j¸×8JÑ8JÐS_×SfÑSfÔgˆJÜ Ÿ;™; {°LÀ*ÓMˆLàÐ%à'¨.Ñ8ˆLà)”r—z‘z bÔ)¨,Ó7ˆð $×(Ñ(¨¯©Ó5ˆØ×(Ñ(¨Ó6ˆð Ð Ø'¨)Ñ3ˆLä—l‘l <°Ó7ˆà˜LÐ(Ð(r}   c                 óN  — |j                  «       \  }}}}	|j                  «       \  }
}
}}
t        j                  ||z  ||t        j                  |j                  ¬«      }d}| j
                  r |t        |j                  d«      «      dz  z  }| j                  r|t        | j                  dz   «      z  }t        d¬«      5  |j                  d||	«      |j                  dd«      j                  d|	|«      }}t        j                  ||j                  «       |j                  «       d	|¬
«      }|j                  ||||«      }d d d «       | j                  s¬|j                  d«      |j                  d«      }}| j                  d d …d d …||z
  |…d |…f   }t        j                  |j                   «      j"                  }t        j$                  ||j                   |j                  ¬«      }t        j&                  |||«      }|�||z   } t)        j*                  d¬«      |«      }|j                   t        j                  k7  rt-        d«      ‚|j/                  |j                   «      }| j1                  |«      }|�||z  }t        j2                  ||«      }||fS # 1 sw Y   �ŒZxY w)Nrº   ç      ð?r,   r¹   r   F)Úenabledr¸   r   )ÚbetaÚalphar¯   zDError with upcasting, attn_weights does not have dtype torch.float32)r¾   rZ   ÚemptyÚfloat32r»   r¢   rŠ   r£   r�   r   r\   r½   Úbaddbmmr�   r4   r¿   r“   rÀ   r~   rÁ   r   rÂ   ÚRuntimeErrorrÃ   r¨   r¼   )r{   rÄ   rÅ   rÆ   rÇ   rÈ   Úbszrž   Ú	q_seq_lenÚdkÚ_Ú	k_seq_lenrÉ   Úscale_factorÚqÚkrÊ   rË   rÌ   rÍ   rÎ   s                        r)   Ú_upcast_and_reordered_attnz,ImageGPTAttention._upcast_and_reordered_attn  sa  € à(-¯
©
«Ñ%ˆˆY˜	 2Ø ŸX™X›ZÑˆˆ1ˆi˜ô —{‘{ 3¨¡?°I¸yÔPU×P]ÑP]Ðfk×frÑfrÔsˆð ˆØ×"Ò"ØœE %§*¡*¨R£.Ó1°SÑ8Ñ8ˆLà×/Ò/ØœE $§.¡.°1Ñ"4Ó5Ñ5ˆLô ˜eÔ$ñ 	VØ—=‘=  Y°Ó3°S·]±]À2ÀrÓ5J×5RÑ5RÐSUÐWYÐ[dÓ5eˆqˆAÜ Ÿ=™=¨°q·w±w³yÀ!Ç'Á'Ã)ÐRSÐ[gÔhˆLØ'×/Ñ/°°YÀ	È9ÓUˆL÷	Vð
 ×&Ò&à',§z¡z°"£~°s·x±xÀ³|˜*ˆLØŸ)™)¢A¢q¨*°|Ñ*CÀjÐ*PÐR]ÐS]ÐR]Ð$]Ñ^ˆKÜŸ™ \×%7Ñ%7Ó8×<Ñ<ˆJô Ÿ™ j¸×8JÑ8JÐS_×SfÑSfÔgˆJÜ Ÿ;™; {°LÀ*ÓMˆLàÐ%à'¨.Ñ8ˆLà)”r—z‘z bÔ)¨,Ó7ˆð ×Ñ¤§¡Ò.ÜÐeÓfÐfØ#×(Ñ(¨¯©Ó5ˆØ×(Ñ(¨Ó6ˆð Ð Ø'¨)Ñ3ˆLä—l‘l <°Ó7ˆà˜LÐ(Ð(÷C	Vñ 	Vús   ÃBJÊJ$c                 óx   — |j                  «       dd ||fz   } |j                  |Ž }|j                  dddd«      S )zJ
        Splits hidden_size dim into attn_head_size and num_heads
        Nr,   r   r?   r   r   )r¾   r›   Úpermute©r{   r~   rž   Úattn_head_sizeÚ	new_shapes        r)   Ú_split_headszImageGPTAttention._split_headsC  sE   € ð —K‘K“M # 2Ð&¨)°^Ð)DÑDˆ	Ø�—‘˜iÐ(ˆØ�~‰~˜a  A qÓ)Ð)r}   c                 óœ   — |j                  dddd«      j                  «       }|j                  «       dd ||z  fz   }|j                  |«      S )zS
        Merges attn_head_size dim and num_attn_heads dim into hidden_size
        r   r?   r   r   Nr¸   )rã   Ú
contiguousr¾   r›   rä   s        r)   Ú_merge_headszImageGPTAttention._merge_headsK  sO   € ð —‘  1 a¨Ó+×6Ñ6Ó8ˆØ—K‘K“M # 2Ð&¨)°nÑ*DÐ)FÑFˆ	Ø�{‰{˜9Ó%Ð%r}   Úhidden_statesÚ
layer_pastrÇ   rÈ   Úencoder_hidden_statesÚencoder_attention_maskÚ	use_cacheÚoutput_attentionsr   c	                 ó´  — |�Zt        | d«      st        d«      ‚| j                  |«      }	| j                  |«      j	                  | j
                  d¬«      \  }
}|}n0| j                  |«      j	                  | j
                  d¬«      \  }	}
}| j                  |	| j                  | j                  «      }	| j                  |
| j                  | j                  «      }
| j                  || j                  | j                  «      }|�7|\  }}t        j                  ||
fd¬«      }
t        j                  ||fd¬«      }|du r|
|f}nd }| j                  r| j                  |	|
|||«      \  }}n| j                  |	|
|||«      \  }}| j                  || j                  | j                  «      }| j                  |«      }| j!                  |«      }||f}|r||fz  }|S )Nr¥   z«If class is used as cross attention, the weights `q_attn` have to be defined. Please make sure to instantiate class with `ImageGPTAttention(..., is_cross_attention=True)`.r?   r¯   r¸   T)Úhasattrr¡   r¥   r:   rP   r    rç   rž   rŸ   rZ   r±   r¤   rá   rÏ   rê   r<   rª   )r{   rë   rì   rÇ   rÈ   rí   rî   rï   rð   rÄ   rÅ   rÆ   Úpast_keyÚ
past_valueÚpresentrÎ   rÉ   Úoutputss                     r)   r†   zImageGPTAttention.forwardS  sÎ  € ð !Ð,Ü˜4 Ô*Ü ðtóð ð
 —K‘K Ó.ˆEØŸ™Ð%:Ó;×AÑAÀ$Ç/Á/ÐWXÐAÓY‰JˆC�Ø3‰Nà $§¡¨MÓ :× @Ñ @ÀÇÁÐVWÐ @Ó XÑˆE�3˜à×!Ñ! %¨¯©¸¿¹ÓGˆØ×Ñ  T§^¡^°T·]±]ÓCˆØ×!Ñ! %¨¯©¸¿¹ÓGˆàÐ!Ø#-Ñ ˆH�jÜ—)‘)˜X s˜O°Ô4ˆCÜ—I‘I˜z¨5Ð1°rÔ:ˆEà˜ÑØ˜E�l‰GàˆGà×'Ò'Ø(,×(GÑ(GÈÈsÐTYÐ[iÐktÓ(uÑ%ˆK™à(,¯
©
°5¸#¸uÀnÐV_Ó(`Ñ%ˆK˜à×'Ñ'¨°T·^±^ÀTÇ]Á]ÓSˆØ—k‘k +Ó.ˆØ×(Ñ(¨Ó5ˆà Ð(ˆÙØ˜�Ñ&ˆGàˆr}   )FN)NN©NNNNNFF)r‡   rˆ   r‰   r   rš   rV   rx   r¶   rÏ   rá   rç   rê   rZ   rz   Útupler†   r‹   rŒ   s   @r)   rŽ   rŽ   ®   så   ø„ ñ)"°8¸D±>ð )"ÐV^Ð_bÑVcõ )"òV;ó$)óL2)òh*ò&ð &*Ø15Ø,0Ø8<Ø9=Ø$)Ø,1ñ3à—|‘|ð3ð ˜T‘Nð3ð ! §¡Ñ.ð	3ð
 ˜EŸL™LÑ)ð3ð  (¨¯©Ñ5ð3ð !)¨¯©Ñ 6ð3ð ˜D‘>ð3ð $ D™>ð3ð 
÷3r}   rŽ   c                   óV   ‡ — e Zd Zˆ fd„Zdej
                  dej
                  fd„Zˆ xZS )ÚImageGPTMLPc                 óö   •— t         ‰| �  «        |j                  }t        ||«      | _        t        ||«      | _        t        |j                     | _        t        j                  |j                  «      | _        y rv   )rw   rx   rs   r   Úc_fcr<   r   Úactivation_functionÚactr   r¦   r©   Údropout)r{   Úintermediate_sizerb   rœ   r|   s       €r)   rx   zImageGPTMLP.__init__Š  s_   ø€ Ü‰ÑÔØ×&Ñ&ˆ	ÜÐ,¨iÓ8ˆŒ	Ü˜YÐ(9Ó:ˆŒÜ˜&×4Ñ4Ñ5ˆŒÜ—z‘z &×"4Ñ"4Ó5ˆ�r}   rë   r   c                 óŽ   — | j                  |«      }| j                  |«      }| j                  |«      }| j                  |«      }|S rv   )rü   rþ   r<   rÿ   )r{   rë   s     r)   r†   zImageGPTMLP.forward’  s@   € ØŸ	™	 -Ó0ˆØŸ™ Ó/ˆØŸ™ MÓ2ˆØŸ™ ]Ó3ˆØÐr}   )r‡   rˆ   r‰   rx   rZ   rz   r†   r‹   rŒ   s   @r)   rú   rú   ‰  s#   ø„ ô6ð U§\¡\ð °e·l±l÷ r}   rú   c                   óê   ‡ — e Zd Zdˆ fd„	Z	 	 	 	 	 	 	 ddej
                  dee   deej
                     deej
                     deej
                     deej
                     dee   d	ee   d
efd„Z	ˆ xZ
S )ÚImageGPTBlockc                 ó   •— t         ‰| �  «        |j                  }|j                  �|j                  nd|z  }t	        ||j
                  ¬«      | _        t        ||¬«      | _        t	        ||j
                  ¬«      | _	        |j                  r/t        |d|¬«      | _        t	        ||j
                  ¬«      | _        t        ||«      | _        y )Né   ©rt   ©r�   T)r�   r�   )rw   rx   rs   Ún_innerrr   Úlayer_norm_epsilonÚln_1rŽ   r;   Úln_2Úadd_cross_attentionÚcrossattentionÚln_cross_attnrú   Úmlp)r{   rb   r�   rs   Ú	inner_dimr|   s        €r)   rx   zImageGPTBlock.__init__›  s¥   ø€ Ü‰ÑÔØ×(Ñ(ˆØ&,§n¡nÐ&@�F—N’NÀaÈ+Áoˆ	ä% k°v×7PÑ7PÔQˆŒ	Ü% f¸	ÔBˆŒ	Ü% k°v×7PÑ7PÔQˆŒ	à×%Ò%Ü"3°FÈtÐ_hÔ"iˆDÔÜ!2°;ÀF×D]ÑD]Ô!^ˆDÔä˜y¨&Ó1ˆ�r}   rë   rì   rÇ   rÈ   rí   rî   rï   rð   r   c	                 óž  — |}	| j                  |«      }| j                  ||||||¬«      }
|
d   }|
dd  }||	z   }|�Wt        | d«      st        d| › d�«      ‚|}	| j	                  |«      }| j                  ||||||¬«      }|d   }|	|z   }||dd  z   }|}	| j                  |«      }| j                  |«      }|	|z   }|f|r|z   }|S |dd  z   }|S )	N)rì   rÇ   rÈ   rï   rð   r   r   r  z'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`)rÇ   rÈ   rí   rî   rð   r?   )r
  r;   rò   r¡   r  r  r  r  )r{   rë   rì   rÇ   rÈ   rí   rî   rï   rð   ÚresidualÚattn_outputsrÎ   rö   Úcross_attn_outputsÚfeed_forward_hidden_statess                  r)   r†   zImageGPTBlock.forwardª  sN  € ð !ˆØŸ	™	 -Ó0ˆØ—y‘yØØ!Ø)ØØØ/ð !ó 
ˆð # 1‘oˆØ˜q˜rÐ"ˆà# hÑ.ˆà Ð,ä˜4Ð!1Ô2Ü Ø=¸d¸Vð DZð Zóð ð %ˆHØ ×.Ñ.¨}Ó=ˆMØ!%×!4Ñ!4ØØ-Ø#Ø&;Ø'=Ø"3ð "5ó "Ðð -¨QÑ/ˆKà$ {Ñ2ˆMØÐ 2°1°2Ð 6Ñ6ˆGà ˆØŸ	™	 -Ó0ˆØ%)§X¡X¨mÓ%<Ð"à Ð#=Ñ=ˆà Ð"± gÑLˆàˆð AHÈÈÀÑLˆàˆr}   rv   r÷   )r‡   rˆ   r‰   rx   rZ   rz   r   rš   rø   r†   r‹   rŒ   s   @r)   r  r  š  s°   ø„ õ2ð$ &*Ø15Ø,0Ø8<Ø9=Ø$)Ø,1ñ8à—|‘|ð8ð ˜T‘Nð8ð ! §¡Ñ.ð	8ð
 ˜EŸL™LÑ)ð8ð  (¨¯©Ñ5ð8ð !)¨¯©Ñ 6ð8ð ˜D‘>ð8ð $ D™>ð8ð 
÷8r}   r  c                   óB   ‡ — e Zd ZdZeZeZdZdZ	dZ
dgZˆ fd„Zd„ Zˆ xZS )ÚImageGPTPreTrainedModelz†
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    r/   Ú	input_idsTr  c                 ó$   •— t        ‰| �  |i |¤Ž y rv   )rw   rx   )r{   ÚinputsÚkwargsr|   s      €r)   rx   z ImageGPTPreTrainedModel.__init__ò  s   ø€ Ü‰Ñ˜&Ð+ FÓ+r}   c           	      ó¬  — t        |t        j                  t        f«      rl|j                  j
                  j                  d| j                  j                  ¬«       |j                  �í|j                  j
                  j                  «        nÈt        |t        j                  «      ry|j                  j
                  j                  d| j                  j                  ¬«       |j                  �g|j                  j
                  |j                     j                  «        n5t        |t        «      r%|j                  j
                  j                  d«       |j                  «       D ]m  \  }}d|v sŒd|v sŒ|j
                  j                  d| j                  j                  t!        j"                  d| j                  j$                  z  «      z  ¬«       Œo y)zInitialize the weights.g        )r„   ÚstdNrÑ   r<   r2   r?   )Ú
isinstancer   ÚLinearr   r2   r_   Únormal_rb   Úinitializer_ranger4   Úzero_Ú	EmbeddingÚpadding_idxrr   Úfill_Únamed_parametersÚmathrƒ   Ún_layer)r{   Úmoduleri   Úps       r)   Ú_init_weightsz%ImageGPTPreTrainedModel._init_weightsõ  sQ  € ä�fœrŸy™y¬&Ð1Ô2ð �M‰M×Ñ×&Ñ&¨C°T·[±[×5RÑ5RÐ&ÔSØ�{‰{Ð&Ø—‘× Ñ ×&Ñ&Õ(Ü˜¤§¡Ô-Ø�M‰M×Ñ×&Ñ&¨C°T·[±[×5RÑ5RÐ&ÔSØ×!Ñ!Ð-Ø—‘×"Ñ" 6×#5Ñ#5Ñ6×<Ñ<Õ>Ü˜Ô 1Ô2Ø�M‰M×Ñ×$Ñ$ SÔ)ð ×.Ñ.Ó0ò 	s‰GˆD�!Ø˜4Ò H°Ò$4à—‘—‘ C¨d¯k©k×.KÑ.KÌdÏiÉiÐXYÐ\`×\gÑ\g×\oÑ\oÑXoÓNpÑ.p�Õrñ	sr}   )r‡   rˆ   r‰   Ú__doc__r   Úconfig_classrp   Úload_tf_weightsÚbase_model_prefixÚmain_input_nameÚsupports_gradient_checkpointingÚ_no_split_modulesrx   r+  r‹   rŒ   s   @r)   r  r  å  s9   ø„ ñð
 "€LØ1€OØ%ÐØ!€OØ&*Ð#Ø(Ð)Ðô,ösr}   r  aB  

    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 ([`ImageGPTConfig`]): 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][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 [`AutoImageProcessor`]. See [`ImageGPTImageProcessor.__call__`] for details.

        past_key_values (`Tuple[Tuple[torch.Tensor]]` of length `config.n_layers`):
            Contains precomputed 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.

            If `past_key_values` is used, optionally only the last `inputs_embeds` have to be input (see
            `past_key_values`).
        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.
zbThe bare ImageGPT Model transformer outputting raw hidden-states without any specific head on top.c            #       óÞ  ‡ — e Zd Zdef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j                     deej                     dee   dee   dee   dee   dedeeef   fd„«       «       Zˆ xZS )ÚImageGPTModelrb   c           	      óv  •— t         ‰| �  |«       |j                  | _        t	        j
                  |j                  | j                  «      | _        t	        j
                  |j                  | j                  «      | _	        t	        j                  |j                  «      | _        t	        j                  t        |j                  «      D �cg c]  }t!        ||¬«      ‘Œ c}«      | _        t%        | j                  |j&                  ¬«      | _        d| _        d | _        d| _        | j1                  «        y c c}w )Nr  r  F)rw   rx   rs   rœ   r   r#  r`   r6   r–   r5   r¦   Ú
embd_pdropÚdropÚ
ModuleListÚrangeÚnum_hidden_layersr  Úhrr   r	  Úln_fÚmodel_parallelÚ
device_mapÚgradient_checkpointingÚ	post_init)r{   rb   Úir|   s      €r)   rx   zImageGPTModel.__init__d  sÚ   ø€ Ü‰Ñ˜Ô à×+Ñ+ˆŒä—<‘< × 1Ñ 1°4·>±>ÓBˆŒÜ—<‘< × >Ñ >ÀÇÁÓOˆŒä—J‘J˜v×0Ñ0Ó1ˆŒ	Ü—‘ÌEÐRX×RjÑRjÓLkÖlÀq¤¨fÀÖ BÒlÓmˆŒÜ% d§n¡n¸&×:SÑ:SÔTˆŒ	ð $ˆÔØˆŒØ&+ˆÔ#à�‰Õùò  ms   Ã
D6c                 ó   — | j                   S rv   ©r6   ©r{   s    r)   Úget_input_embeddingsz"ImageGPTModel.get_input_embeddingsw  s   € Ø�x‰xˆr}   c                 ó   — || _         y rv   rC  ©r{   Únew_embeddingss     r)   Úset_input_embeddingsz"ImageGPTModel.set_input_embeddingsz  s	   € Ø!ˆ�r}   c                 ó„   — |j                  «       D ]-  \  }}| j                  |   j                  j                  |«       Œ/ y)zv
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
        N)Úitemsr;  r;   r¶   )r{   Úheads_to_pruneÚlayerr³   s       r)   Ú_prune_headszImageGPTModel._prune_heads}  s<   € ð +×0Ñ0Ó2ò 	2‰LˆE�5Ø�F‰F�5‰M×Ñ×*Ñ*¨5Õ1ñ	2r}   ©Úoutput_typer-  r  Úpast_key_valuesrÇ   Útoken_type_idsÚposition_idsrÈ   Úinputs_embedsrí   rî   rï   rð   Úoutput_hidden_statesÚreturn_dictr  r   c                 óÐ  ‡$— d|v r8t        j                  dt        «       |�t        d«      ‚|j	                  d«      }|�|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                  }|�|j                  d|d   «      }|€%d}t        dgt!        | j"                  «      z  «      }n|d   d   j                  d	«      }|€>t%        j&                  ||d   |z   t$        j(                  |¬
«      }|j+                  d«      }|�z|dk  rt        d«      ‚|j                  |d«      }|dd…dddd…f   }|j-                  | j.                  ¬«      }d|z
  t%        j0                  | j.                  «      j2                  z  }| j
                  j4                  rE|�C|j                  «       \  }}}||f}|	€t%        j6                  ||¬«      }	| j9                  |	«      }	nd}	| j;                  || j
                  j<                  «      }|€| j?                  |«      }| jA                  |«      }||j-                  |j                  «      z   Š$|�| j?                  |«      }‰$|z   Š$| jC                  ‰$«      Š$|‰$j                  d«      fz   }| jD                  r%| jF                  r|
rtH        jK                  d«       d}
|
rdnd}|rdnd}|r| j
                  j4                  rdnd}|rdnd}tM        tO        | j"                  |«      «      D �]¹  \  }\  }} | jP                  r‘t$        jR                  jU                  ‰$j                  «       | �t        ˆ$fd„| D «       «      } |�|j-                  ‰$j                  «      }tW        |t$        jX                  «      r|j-                  ‰$j                  «      }|r|‰$fz   }| jD                  r3| jF                  r'| j[                  |j\                  ‰$d|||   ||	|
|«	      }!n |‰$| |||   ||	|
|¬«      }!|!d   Š$|
du r	||!d   fz   }|r0||!|
rdnd   fz   }| j
                  j4                  r||!|
rdnd   fz   }| jP                  s�ŒS| j^                  ja                  «       D ]J  \  }"}#||#d   k(  sŒdtc        |"«      z   | jd                  k7  sŒ+‰$j-                  dtc        |"dz   «      z   «      Š$ŒL �Œ¼ | jg                  ‰$«      Š$ ‰$j                  |Ž Š$|r|‰$fz   }|st        d„ ‰$||||fD «       «      S ti        ‰$||||¬«      S )aR  
        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:

        Examples:

        ```python
        >>> from transformers import AutoImageProcessor, ImageGPTModel
        >>> from PIL import Image
        >>> import requests

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> image_processor = AutoImageProcessor.from_pretrained("openai/imagegpt-small")
        >>> model = ImageGPTModel.from_pretrained("openai/imagegpt-small")

        >>> inputs = image_processor(images=image, return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> last_hidden_states = outputs.last_hidden_state
        ```Úpixel_valuesú`The `pixel_values` argument is deprecated and will be removed in v4.47, use `input_ids` instead.Nú_You cannot pass both `pixel_values` and `input_ids`. Please make sure to only pass `input_ids`.zDYou cannot specify both input_ids and inputs_embeds at the same timer,   r   z5You have to specify either input_ids or inputs_embedsr¸   rº   z$batch_size has to be defined and > 0r’   rÑ   )r»   zZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...Fr&   c              3   óT   •K  — | ]  }|j                  ‰j                  «      –— Œ! y ­wrv   )Útor»   )r'   Ú
past_staterë   s     €r)   r*   z(ImageGPTModel.forward.<locals>.<genexpr>#  s    øè ø€ Ò&hÈz z§}¡}°]×5IÑ5I×'JÑ&hùs   ƒ%()rì   rÇ   rÈ   rí   rî   rï   rð   Tr   r?   r   zcuda:c              3   ó$   K  — | ]  }|�|–— Œ
 y ­wrv   r&   )r'   Úvs     r)   r*   z(ImageGPTModel.forward.<locals>.<genexpr>[  s   è ø€ ò àØ�=ô ñùr+   )Úlast_hidden_staterQ  rë   Ú
attentionsÚcross_attentions)5ÚwarningsÚwarnÚFutureWarningr¡   Úpoprb   rð   rU  rï   Úuse_return_dictÚ%warn_if_padding_and_no_attention_maskr¾   r›   rW   r»   rø   rU   r;  rZ   ÚarangeÚlongÚ	unsqueezer\  r“   r¿   rÀ   r  r™   Úinvert_attention_maskÚget_head_maskr(  r6   r5   r7  r?  ÚtrainingrC   Úwarning_onceÚ	enumeraterO   r=  ÚcudaÚ
set_devicer  rz   Ú_gradient_checkpointing_funcÚ__call__r>  rK  ÚstrÚlast_devicer<  r   )%r{   r  rQ  rÇ   rR  rS  rÈ   rT  rí   rî   rï   rð   rU  rV  r  Úinput_shapeÚ
batch_sizer»   Úpast_lengthÚencoder_batch_sizeÚencoder_sequence_lengthrÜ   Úencoder_hidden_shapeÚposition_embedsÚtoken_type_embedsÚoutput_shapeÚpresentsÚall_self_attentionsÚall_cross_attentionsÚall_hidden_statesrA  Úblockrì   rö   rà   r_  rë   s%                                       @r)   r†   zImageGPTModel.forward„  s8  ø€ ðZ ˜VÑ#Ü�M‰MØrÜôð
 Ð$Ü Øuóð ð Ÿ
™
 >Ó2ˆIà1BÐ1NÑ-ÐTX×T_ÑT_×TqÑTqÐà$8Ð$DÑ È$Ï+É+×JjÑJjð 	ð "+Ð!6‘I¸D¿K¹K×<QÑ<Qˆ	Ø%0Ð%<‘kÀ$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+×0Ñ0°°[À±_ÓEˆNàÐ"ØˆKÜ# T F¬S°·±«[Ñ$8Ó9‰Oà)¨!Ñ,¨QÑ/×4Ñ4°RÓ8ˆKØÐÜ Ÿ<™<¨°[À±_À{Ñ5RÔZ_×ZdÑZdÐmsÔtˆLØ'×1Ñ1°!Ó4ˆLð Ð%Ø˜QŠÜ Ð!GÓHÐHØ+×0Ñ0°¸RÓ@ˆNð ,ªA¨t°Tº1Ð,<Ñ=ˆNð ,×.Ñ.°T·Z±ZÐ.Ó@ˆNØ! NÑ2´e·k±kÀ$Ç*Á*Ó6M×6QÑ6QÑQˆNð �;‰;×*Ò*Ð/DÐ/PØ=R×=WÑ=WÓ=YÑ:ÐÐ 7¸Ø$6Ð8OÐ#PÐ Ø%Ð-Ü).¯©Ð4HÐQWÔ)XÐ&Ø%)×%?Ñ%?Ð@VÓ%WÑ"à%)Ð"ð ×&Ñ& y°$·+±+×2EÑ2EÓFˆ	àÐ Ø ŸH™H YÓ/ˆMØŸ(™( <Ó0ˆØ%¨×(:Ñ(:¸=×;OÑ;OÓ(PÑPˆàÐ%Ø $§¡¨Ó 8ÐØ)Ð,=Ñ=ˆMàŸ	™	 -Ó0ˆà" m×&8Ñ&8¸Ó&<Ð%>Ñ>ˆà×&Ò&¨4¯=ª=ÙÜ×#Ñ#Øpôð "�	á"‘2¨ˆÙ$5™b¸4ÐÙ%6¸4¿;¹;×;ZÒ;Z™rÐ`dÐÙ"6™B¸DÐÜ&/´°D·F±F¸OÓ0LÓ&Mó 4	OÑ"ˆAÑ"��zà×"Ò"Ü—
‘
×%Ñ% m×&:Ñ&:Ô;àÐ)Ü!&Ó&hÐ]gÔ&hÓ!h�Jà!Ð-Ø%3×%6Ñ%6°}×7KÑ7KÓ%L�NÜ˜i¬¯©Ô6Ø )§¡¨]×-AÑ-AÓ B�IÙ#Ø$5¸Ð8HÑ$HÐ!à×*Ò*¨t¯}ª}Ø×;Ñ;Ø—N‘NØ!ØØ"Ø˜a‘LØ)Ø*ØØ%ó
‘ñ  Ø!Ø)Ø#1Ø'¨™lØ*?Ø+AØ'Ø&7ô	�ð $ A™JˆMØ˜DÑ Ø# w¨q¡z mÑ3�á Ø&9¸WÉ)ÁQÐYZÑ=[Ð<]Ñ&]Ð#Ø—;‘;×2Ò2Ø+?À7ÑPYÉ1Ð_`ÑCaÐBcÑ+cÐ(ð ×"Ô"Ø ŸO™O×1Ñ1Ó3ò O‘D�A�qØ˜A˜b™E“z g´°A³Ñ&6¸$×:JÑ:JÓ&JØ(5×(8Ñ(8¸Ä3ÀqÈ1ÁuÃ:Ñ9MÓ(N™òOðe4	Oðl Ÿ	™	 -Ó0ˆà*˜×*Ñ*¨LÐ9ˆáØ 1°]Ð4DÑ DÐáÜñ à'¨Ð3DÐFYÐ[oÐpôó ð ô 9Ø+Ø$Ø+Ø*Ø1ô
ð 	
r}   )NNNNNNNNNNNNN)r‡   rˆ   r‰   r   rx   rE  rI  rN  r   ÚIMAGEGPT_INPUTS_DOCSTRINGr   r   Ú_CONFIG_FOR_DOCr   rZ   rz   r   rš   r   r   r†   r‹   rŒ   s   @r)   r4  r4  _  s�  ø„ ð
˜~õ ò&ò"ò2ñ +Ð+DÓEÙÐ+TÐcrÔsð -1Ø@DØ15Ø15Ø/3Ø,0Ø04Ø8<Ø9=Ø$(Ø,0Ø/3Ø&*ña
à˜EŸL™LÑ)ða
ð " %¨¨e¯l©lÑ(;Ñ"<Ñ=ða
ð ! §¡Ñ.ð	a
ð
 ! §¡Ñ.ða
ð ˜uŸ|™|Ñ,ða
ð ˜EŸL™LÑ)ða
ð   §¡Ñ-ða
ð  (¨¯©Ñ5ða
ð !)¨¯©Ñ 6ða
ð ˜D‘>ða
ð $ D™>ða
ð ' t™nða
ð ˜d‘^ða
ð ða
ð  
ˆuÐ?Ð?Ñ	@ò!a
ó tó Fôa
r}   r4  z‹
    The ImageGPT Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    c            %       óp  ‡ — e Zd ZdgZdef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j                     deej                     dee   dee   dee   dee   dedeeef   f d„«       «       Zedeeej                        dej                  deeej                        fd„«       Zˆ xZS )ÚImageGPTForCausalImageModelingzlm_head.weightrb   c                 óè   •— t         ‰| �  |«       t        |«      | _        t	        j
                  |j                  |j                  dz
  d¬«      | _        d| _	        d | _
        | j                  «        y )Nr   F©r4   )rw   rx   r4  r/   r   r  r]   r`   r=   r=  r>  r@  ©r{   rb   r|   s     €r)   rx   z'ImageGPTForCausalImageModeling.__init__t  s[   ø€ Ü‰Ñ˜Ô Ü(¨Ó0ˆÔÜ—y‘y §¡°×0AÑ0AÀAÑ0EÈEÔRˆŒð $ˆÔØˆŒà�‰Õr}   c                 ó   — | j                   S rv   ©r=   rD  s    r)   Úget_output_embeddingsz4ImageGPTForCausalImageModeling.get_output_embeddings  s   € Ø�|‰|Ðr}   c                 ó   — || _         y rv   r�  rG  s     r)   Úset_output_embeddingsz4ImageGPTForCausalImageModeling.set_output_embeddings‚  s	   € Ø%ˆ�r}   rO  r  rQ  rÇ   rR  rS  rÈ   rT  rí   rî   Úlabelsrï   rð   rU  rV  r  r   c                 óž  — d|v r8t        j                  dt        «       |�t        d«      ‚|j	                  d«      }|�|n| j
                  j                  }| j                  |||||||||	||||¬«      }|d   }| j                  |«      }d}|
�r|ddd…dd…f   j                  «       }|
dd	d…f   j                  «       }t        «       } ||j                  d|j                  d«      «      |j                  d«      «      }|s|f|d	d z   }|�|f|z   S |S t        |||j                  |j                  |j                   |j"                  ¬
«      S )aã  
        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:

        Examples:

        ```python
        >>> from transformers import AutoImageProcessor, ImageGPTForCausalImageModeling
        >>> import torch
        >>> import matplotlib.pyplot as plt
        >>> import numpy as np

        >>> image_processor = AutoImageProcessor.from_pretrained("openai/imagegpt-small")
        >>> model = ImageGPTForCausalImageModeling.from_pretrained("openai/imagegpt-small")
        >>> device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        >>> model.to(device)  # doctest: +IGNORE_RESULT

        >>> # unconditional generation of 8 images
        >>> batch_size = 4
        >>> context = torch.full((batch_size, 1), model.config.vocab_size - 1)  # initialize with SOS token
        >>> context = context.to(device)
        >>> output = model.generate(
        ...     input_ids=context, max_length=model.config.n_positions + 1, temperature=1.0, do_sample=True, top_k=40
        ... )

        >>> clusters = image_processor.clusters
        >>> height = image_processor.size["height"]
        >>> width = image_processor.size["width"]

        >>> samples = output[:, 1:].detach().cpu().numpy()
        >>> samples_img = [
        ...     np.reshape(np.rint(127.5 * (clusters[s] + 1.0)), [height, width, 3]).astype(np.uint8) for s in samples
        ... ]  # convert color cluster tokens back to pixels
        >>> f, axes = plt.subplots(1, batch_size, dpi=300)

        >>> for img, ax in zip(samples_img, axes):  # doctest: +IGNORE_RESULT
        ...     ax.axis("off")
        ...     ax.imshow(img)
        ```rX  rY  NrZ  )rQ  rÇ   rR  rS  rÈ   rT  rí   rî   rï   rð   rU  rV  r   .r,   r   )ÚlossÚlogitsrQ  rë   ra  rb  )rc  rd  re  r¡   rf  rb   rg  r/   r=   ré   r
   r›   r¾   r   rQ  rë   ra  rb  )r{   r  rQ  rÇ   rR  rS  rÈ   rT  rí   rî   r‘  rï   rð   rU  rV  r  Útransformer_outputsrë   Ú	lm_logitsr“  Úshift_logitsÚshift_labelsÚloss_fctÚoutputs                           r)   r†   z&ImageGPTForCausalImageModeling.forward…  s‘  € ð@ ˜VÑ#Ü�M‰MØrÜôð
 Ð$Ü Øuóð ð Ÿ
™
 >Ó2ˆIà%0Ð%<‘kÀ$Ç+Á+×B]ÑB]ˆà"×.Ñ.ØØ+Ø)Ø)Ø%ØØ'Ø"7Ø#9ØØ/Ø!5Ø#ð /ó 
Ðð ,¨AÑ.ˆà—L‘L Ó/ˆ	àˆØÐà$ S¨#¨2¨#ªq [Ñ1×<Ñ<Ó>ˆLØ! # q¡r '™?×5Ñ5Ó7ˆLä'Ó)ˆHÙ˜L×-Ñ-¨b°,×2CÑ2CÀBÓ2GÓHÈ,×J[ÑJ[Ð\^ÓJ_Ó`ˆDáØ�\Ð$7¸¸Ð$;Ñ;ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä0ØØØ/×?Ñ?Ø-×;Ñ;Ø*×5Ñ5Ø0×AÑAô
ð 	
r}   Úbeam_idxc                 ó,   ‡— t        ˆfd„| D «       «      S )a  
        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
        beam_idx at every generation step.
        c              3   óF   •K  — | ]  }t        ˆfd „|D «       «      –— Œ y­w)c              3   ót   •K  — | ]/  }|j                  d ‰j                  |j                  «      «      –— Œ1 y­w)r   N)Úindex_selectr\  r»   )r'   r]  r›  s     €r)   r*   zJImageGPTForCausalImageModeling._reorder_cache.<locals>.<genexpr>.<genexpr>  s.   øè ø€ ÒjÐQ[�*×)Ñ)¨!¨X¯[©[¸×9JÑ9JÓ-K×LÑjùs   ƒ58N©rø   )r'   rì   r›  s     €r)   r*   z@ImageGPTForCausalImageModeling._reorder_cache.<locals>.<genexpr>  s%   øè ø€ ò 
àô ÓjÐ_iÔj×jñ
ùs   ƒ!r   )rQ  r›  s    `r)   Ú_reorder_cachez-ImageGPTForCausalImageModeling._reorder_cacheý  s   ø€ ô ó 
à-ô
ó 
ð 	
r}   )NNNNNNNNNNNNNN)r‡   rˆ   r‰   Ú_tied_weights_keysr   rx   rŽ  r�  r   r…  r   r   r†  r   rZ   rz   r   rš   r   r   r†   Ústaticmethodr¡  r‹   rŒ   s   @r)   rˆ  rˆ  j  sù  ø„ ð +Ð+Ðð	˜~õ 	òò&ñ +Ð+DÓEÙÐ+LÐ[jÔkð -1Ø@DØ15Ø15Ø/3Ø,0Ø04Ø8<Ø9=Ø)-Ø$(Ø,0Ø/3Ø&*ñt
à˜EŸL™LÑ)ðt
ð " %¨¨e¯l©lÑ(;Ñ"<Ñ=ðt
ð ! §¡Ñ.ð	t
ð
 ! §¡Ñ.ðt
ð ˜uŸ|™|Ñ,ðt
ð ˜EŸL™LÑ)ðt
ð   §¡Ñ-ðt
ð  (¨¯©Ñ5ðt
ð !)¨¯©Ñ 6ðt
ð ˜Ÿ™Ñ&ðt
ð ˜D‘>ðt
ð $ D™>ðt
ð ' t™nðt
ð ˜d‘^ðt
ð  ð!t
ð" 
ˆuÐ7Ð7Ñ	8ò#t
ó ló Fðt
ðl ð
Ø˜u U§\¡\Ñ2Ñ3ð
Ø?D¿|¹|ð
à	ˆu�U—\‘\Ñ"Ñ	#ò
ó ô
r}   rˆ  zË
    The ImageGPT Model transformer with an image classification head on top (linear layer).
    [`ImageGPTForImageClassification`] average-pools the hidden states in order to do the classification.
    c            !       ó¬  ‡ — e Zd Zdef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deeef   fd„«       «       Zˆ xZS )ÚImageGPTForImageClassificationrb   c                 óè   •— t         ‰| �  |«       |j                  | _        t        |«      | _        t        j                  |j                  | j                  d¬«      | _        | j                  «        y )NFrŠ  )
rw   rx   Ú
num_labelsr4  r/   r   r  r]   Úscorer@  r‹  s     €r)   rx   z'ImageGPTForImageClassification.__init__  sR   ø€ Ü‰Ñ˜Ô Ø ×+Ñ+ˆŒÜ(¨Ó0ˆÔÜ—Y‘Y˜vŸ}™}¨d¯o©oÀEÔJˆŒ
ð 	�‰Õr}   rO  r  rQ  rÇ   rR  rS  rÈ   rT  r‘  rï   rð   rU  rV  r  r   c                 óÔ  — d|v r8t        j                  dt        «       |�t        d«      ‚|j	                  d«      }|�|n| j
                  j                  }| j                  ||||||||	|
||¬«      }|d   }|j                  d¬«      }| j                  |«      }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=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.                  |j0                  ¬«      S )a7  
        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:

        Examples:

        ```python
        >>> from transformers import AutoImageProcessor, ImageGPTForImageClassification
        >>> from PIL import Image
        >>> import requests

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> image_processor = AutoImageProcessor.from_pretrained("openai/imagegpt-small")
        >>> model = ImageGPTForImageClassification.from_pretrained("openai/imagegpt-small")

        >>> inputs = image_processor(images=image, return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> logits = outputs.logits
        ```rX  rY  NrZ  )
rQ  rÇ   rR  rS  rÈ   rT  rï   rð   rU  rV  r   r   r¯   Ú
regressionÚsingle_label_classificationÚmulti_label_classificationr,   )r“  r”  rQ  rë   ra  )rc  rd  re  r¡   rf  rb   rg  r/   r„   r¨  Úproblem_typer§  r“   rZ   rj  rV   r   rN   r
   r›   r	   r   rQ  rë   ra  )r{   r  rQ  rÇ   rR  rS  rÈ   rT  r‘  rï   rð   rU  rV  r  r•  rë   Úpooled_hidden_statesr”  r“  r™  rš  s                        r)   r†   z&ImageGPTForImageClassification.forward  sG  € ðX ˜VÑ#Ü�M‰MØrÜôð
 Ð$Ü Øuóð ð Ÿ
™
 >Ó2ˆIà%0Ð%<‘kÀ$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,×1Ñ1°aÐ1Ó8Ðà—‘Ð0Ó1ˆ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# F§N¡NÓ$4°f·n±nÓ6FÓG‘Dá# F¨FÓ3‘DØ—‘×)Ñ)Ð-JÒJÜ+Ó-�Ù §¡¨B°·±Ó @À&Ç+Á+ÈbÃ/ÓR‘Ø—‘×)Ñ)Ð-IÒIÜ,Ó.�Ù ¨Ó/�ÙØ�YÐ!4°Q°RÐ!8Ñ8ˆFØ)-Ð)9�T�G˜fÑ$ÐE¸vÐEä/ØØØ/×?Ñ?Ø-×;Ñ;Ø*×5Ñ5ô
ð 	
r}   )NNNNNNNNNNNN)r‡   rˆ   r‰   r   rx   r   r…  r   r   r†  r   rZ   rz   r   rš   r   r   r†   r‹   rŒ   s   @r)   r¥  r¥    si  ø„ ð˜~õ ñ +Ð+DÓEÙÐ+KÐZiÔjð -1Ø@DØ15Ø15Ø/3Ø,0Ø04Ø)-Ø$(Ø,0Ø/3Ø&*ñl
à˜EŸL™LÑ)ðl
ð " %¨¨e¯l©lÑ(;Ñ"<Ñ=ðl
ð ! §¡Ñ.ð	l
ð
 ! §¡Ñ.ðl
ð ˜uŸ|™|Ñ,ðl
ð ˜EŸL™LÑ)ðl
ð   §¡Ñ-ðl
ð ˜Ÿ™Ñ&ðl
ð ˜D‘>ðl
ð $ D™>ðl
ð ' t™nðl
ð ˜d‘^ðl
ð ðl
ð 
ˆuÐ6Ð6Ñ	7òl
ó kó Fôl
r}   r¥  )rˆ  r¥  r4  r  rp   ):r,  r'  rE   rc  Útypingr   r   r   r   rZ   Útorch.utils.checkpointr   Útorch.cuda.ampr   Útorch.nnr	   r
   r   Úactivationsr   Ú
generationr   Úmodeling_outputsr   r   r   Úmodeling_utilsr   Úpytorch_utilsr   r   r   Úutilsr   r   r   r   r   Úconfiguration_imagegptr   Ú
get_loggerr‡   rC   Ú_CHECKPOINT_FOR_DOCr†  rp   ÚModulerr   rŽ   rú   r  r  ÚIMAGEGPT_START_DOCSTRINGr…  r4  rˆ  r¥  Ú__all__r&   r}   r)   ú<module>r¿     ss  ðñ %ã Û 	Û ß .Ó .ã Û Ý Ý #ß AÑ Aå !Ý )÷ñ õ
 .ß YÑ Y÷õ õ 3ð 
ˆ×	Ñ	˜HÓ	%€à-Ð Ø"€òiôX
˜Ÿ	™	ô 
ôX˜Ÿ	™	ô Xôv�"—)‘)ô ô"H�B—I‘Iô HôV(s˜oô (sðVÐ ð <Ð ñ~ ØhØóôD
Ð+ó D
ó	ðD
ñN ðð óôX
Ð%<¸oó X
óðX
ñv ðð óôx
Ð%<ó x
óðx
òv�r}   