
    qi                        d Z ddlZddlmZ ddl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 dd	lmZ dd
lmZ ddlmZmZ ddlmZ ddlmZ ddlmZmZmZmZm Z m!Z! ddl"m#Z#m$Z$m%Z%  e jL                  e'      Z(de	jR                  de	jR                  fdZ*de	jR                  de	jR                  fdZ+de	jR                  de,fdZ-dFde	jR                  de.de/de,de	jR                  f
dZ0dGdZ1d Z2 G d de
jf                        Z4 G d d e
jf                        Z5 G d! d"e
jf                        Z6ee G d# d$e                    Z7 G d% d&e
jf                        Z8 G d' d(e
jf                        Z9 G d) d*e
jf                        Z: G d+ d,e
jf                        Z; G d- d.e
jf                        Z< G d/ d0e<      Z= G d1 d2e
jf                        Z> G d3 d4e      Z?e G d5 d6e             Z@ G d7 d8e
jf                        ZA G d9 d:e
jf                        ZB G d; d<e@      ZC G d= d>e@      ZD G d? d@e
jf                        ZE G dA dBe@      ZFe G dC dDe@             ZGg dEZHy)HzPyTorch GroupViT model.    N)	dataclass)Any)nn   )initialization)ACT2FN)create_causal_mask)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPooling)PreTrainedModel)Unpack)ModelOutputTransformersKwargsauto_docstringcan_return_tuplelogging	torch_int   )GroupViTConfigGroupViTTextConfigGroupViTVisionConfiglogitsreturnc                     t         j                  j                  | t        j                  t        |       | j                              S )Ndevice)r   
functionalcross_entropytorcharangelenr   )r   s    `/opt/pipecat/venv/lib/python3.12/site-packages/transformers/models/groupvit/modeling_groupvit.pycontrastive_lossr$   (   s/    ==&&vu||CKPVP]P]/^__    
similarityc                 Z    t        |       }t        | j                               }||z   dz  S )Ng       @)r$   t)r&   caption_loss
image_losss      r#   groupvit_lossr+   -   s,    #J/L!*,,.1J:%,,r%   dimc                     | j                  |      }|j                  |d      d   }t        j                  | t        j                        j                  ||d      }||j                         z
  |z   }|S )NTkeepdimr   memory_format      ?)softmaxmaxr    
zeros_likelegacy_contiguous_formatscatter_detach)r   r,   y_softindexy_hardrets         r#   hard_softmaxr=   3   sk    ^^C FJJsDJ)!,EfE4R4RS\\]`bgilmF
6==?
"V
+CJr%   tauhardc                 :   t         j                  j                  j                  t        j                  d| j
                  | j                        t        j                  d| j
                  | j                              }|j                  | j                        }| |z   |z  }|j                  |      }|rd|j                  |d      d   }t        j                  | t         j                        j                  ||d      }||j                         z
  |z   }	|	S |}	|	S )N        )r   dtyper2   Tr.   r   r0   )r    distributionsgumbelGumbeltensorr   rB   sampleshaper3   r4   r5   r6   r7   r8   )
r   r>   r?   r,   gumbel_distgumbelsr9   r:   r;   r<   s
             r#   gumbel_softmaxrK   =   s    %%,,33SfllCSfllCK   .G3&G__S!F

3
-a0!!&8V8VW``adfkmpqv}}&/ J Jr%   c                    ||z  | j                   d   z  dz  }||kD  r4t        t        j                  ||z              }| j                   d   |z  }n3t        t        j                  ||z              }| j                   d   |z  }| j                   d   }| j                   d   }| j	                  ||||      } t
        j                  j                  | ||fd|      } | S )a  
    Args:
        attentions (`torch.Tensor`): attention map of shape [batch_size, groups, feat_height*feat_width]
        height (`int`): height of the output attention map
        width (`int`): width of the output attention map
        align_corners (`bool`, *optional*): the `align_corner` argument for `nn.functional.interpolate`.

    Returns:
        `torch.Tensor`: resized attention map of shape [batch_size, groups, height, width]
             ?r   r   bilinearsizemodealign_corners)rH   intnproundreshaper   r   interpolate)	
attentionsheightwidthrS   scale
feat_widthfeat_height
batch_sizegroupss	            r#   resize_attention_mapra   S   s     e^z//22s:E~%%-01
 &&q)Z7"((6E>23%%a(K7
!!!$Ja F##JZPJ**&%z + J r%   c           	      H   g }t        j                         5  d}| D ]i  }|j                  ddd      j                         }||}n||z  }t	        |j                  ddd      j                         g| }|j                  |       k 	 ddd       |d   }|S # 1 sw Y   xY w)a1  
    Args:
        attentions (`tuple(torch.FloatTensor)`: tuple of attention maps returned by `GroupViTVisionTransformer`
        hw_shape (`tuple(int)`): height and width of the output attention map
    Returns:
        `torch.Tensor`: the attention map of shape [batch_size, groups, height, width]
    Nr   rM   r   )r    no_gradpermute
contiguousra   append)rY   hw_shape	attn_mapsprev_attn_masks
attn_maskscur_attn_mapfinal_groupings          r#   get_grouping_from_attentionsrn   q   s     I	 +$ 		+J#++Aq!4??AJ&","1J">/0G0G1a0P0[0[0]i`hiL\*		++ r]N!+ +s   A1BB!c                   *     e Zd Zdef fdZd Z xZS )GroupViTCrossAttentionLayerconfigc                 "   t         |           t        |      | _        t	        j
                  |j                  |j                        | _        t        |      | _
        t	        j
                  |j                  |j                        | _        y Neps)super__init__GroupViTAttentionattnr   	LayerNormhidden_sizelayer_norm_epsnorm2GroupViTMLPmlp	norm_postselfrq   	__class__s     r#   rw   z$GroupViTCrossAttentionLayer.__init__   sb    %f-	\\&"4"4&:O:OP
v&f&8&8f>S>STr%   c                     |}|| j                  ||      d   z   }|| j                  | j                  |            z   }| j                  |      }|S )N)encoder_hidden_statesr   )ry   r   r}   r   )r   querykeyxs       r#   forwardz#GroupViTCrossAttentionLayer.forward   sQ    		%s	;A>>A''NN1r%   )__name__
__module____qualname__r   rw   r   __classcell__r   s   @r#   rp   rp      s    U3 Ur%   rp   c                   2     e Zd Zdef fdZddZd Z xZS )GroupViTAssignAttentionrq   c                    t         |           |j                  dz  | _        t	        j
                  |j                  |j                        | _        t	        j
                  |j                  |j                        | _        t	        j
                  |j                  |j                        | _        t	        j
                  |j                  |j                        | _	        |j                  | _
        y )N      )rv   rw   r{   r\   r   Linearq_projk_projv_projproj
assign_epsr   s     r#   rw   z GroupViTAssignAttention.__init__   s    ''-
ii 2 2F4F4FGii 2 2F4F4FGii 2 2F4F4FGIIf00&2D2DE	 ++r%   c                     |r| j                   rt        |d|      }|S |rt        |d      }|S t        j                  j                  |d      }|S )N)r,   r?   r,   )trainingrK   r=   r   r   r3   )r   ry   rD   r?   s       r#   get_attnz GroupViTAssignAttention.get_attn   sX    dmm!$BT:D  #Db1  }},,Tr,:r%   c                 t   |}| j                  |      }| j                  |      }| j                  |      }||j                  dd      z  | j                  z  }| j                  |      }| j                  |dd      }||j                  dd      | j                  z   z  }||z  }| j                  |      }||fS )Nr   rc   F)rD   r?   Tr,   r/   )	r   r   r   	transposer\   r   sumr   r   )r   r   r   valueraw_attnry   	soft_attnouts           r#   r   zGroupViTAssignAttention.forward   s    E" kk# E" CMM"b11TZZ?}}X&MM(5uME	txxBx5GHUliinI~r%   )TT)r   r   r   r   rw   r   r   r   r   s   @r#   r   r      s    ,3 ,	r%   r   c                   0     e Zd Zdef fdZd Zd Z xZS )GroupViTTokenAssignrq   c                 d   t         |           || _        t        j                  |j
                  |j                        | _        t        |j                  t        j                  j                        r|j                  n|j                  |j                  f}|D cg c]  }t        ||j
                  z         c}\  }}t        ||||      | _        t        j                  |j
                  |j                        | _        t        j                  |j
                  |j                        | _        t%        |      | _        t)        |      | _        t        j                  |j
                  |j                        | _        t/        ||j
                  ||j
                        | _        y c c}w rs   )rv   rw   num_output_groupr   rz   r{   r|   norm_tokens
isinstanceassign_mlp_ratiocollectionsabcIterablerT   GroupViTMixerMLP	mlp_internorm_post_tokensnorm_xrp   pre_assign_attnr   assign
norm_new_xr~   mlp_channels)	r   rq   num_group_tokenr   r   r   
tokens_dimchannels_dimr   s	           r#   rw   zGroupViTTokenAssign.__init__   sK    0<<(:(:@U@UV &11;??3K3KL ##))6+B+BC 	
 JZ#ZACF,>,>(>$?#Z 
L)&/:O_` "V-?-?VEZEZ [ll6#5#56;P;PQ:6B-f5,,v'9'9v?T?TU'0B0BLRXRdRde $[s   F-c                 J    | j                  |      }| j                  |      }|S )z
        Args:
            group_tokens (torch.Tensor): group tokens, [batch_size, num_group_tokens, channels]

        Returns:
            projected_group_tokens (torch.Tensor): [batch_size, num_output_groups, channels]
        )r   r   )r   group_tokensprojected_group_tokenss      r#   project_group_tokenz'GroupViTTokenAssign.project_group_token   s+     "&!=!%!6!67M!N%%r%   c                    | j                  |      }| j                  |      }| j                  |      }| j                  ||      }| j	                  ||      \  }}||z  }|| j                  | j                  |            z   }||fS )z
        Args:
            image_tokens (`torch.Tensor`): image tokens, of shape [batch_size, input_length, channels]
            group_tokens (`torch.Tensor`): group tokens, [batch_size, num_group_tokens, channels]
        )r   r   r   r   r   r   r   )r   image_tokensr   r   new_image_tokens	attentions         r#   r   zGroupViTTokenAssign.forward   s     ''5{{<0!%!9!9,!G!%!5!56Ll![&*kk2H,&W#)22+d.?.?P`@a.bb**r%   )r   r   r   r   rw   r   r   r   r   s   @r#   r   r      s    f3 f*&+r%   r   c                   0   e Zd ZU dZdZej                  dz  ed<   dZej                  dz  ed<   dZ	ej                  dz  ed<   dZ
ej                  dz  ed<   dZej                  dz  ed<   dZej                  dz  ed<   dZeed	<   dZeed
<   dee   fdZy)GroupViTModelOutputa  
    loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `return_loss` is `True`):
        Contrastive loss for image-text similarity.
    logits_per_image (`torch.FloatTensor` of shape `(image_batch_size, text_batch_size)`):
        The scaled dot product scores between `image_embeds` and `text_embeds`. This represents the image-text
        similarity scores.
    logits_per_text (`torch.FloatTensor` of shape `(text_batch_size, image_batch_size)`):
        The scaled dot product scores between `text_embeds` and `image_embeds`. This represents the text-image
        similarity scores.
    segmentation_logits (`torch.FloatTensor` of shape `(batch_size, config.num_labels, logits_height, logits_width)`):
        Classification scores for each pixel.

        <Tip warning={true}>

        The logits returned do not necessarily have the same size as the `pixel_values` passed as inputs. This is
        to avoid doing two interpolations and lose some quality when a user needs to resize the logits to the
        original image size as post-processing. You should always check your logits shape and resize as needed.

        </Tip>
    text_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim`):
        The text embeddings obtained by applying the projection layer to the pooled output of
        [`GroupViTTextModel`].
    image_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim`):
        The image embeddings obtained by applying the projection layer to the pooled output of
        [`GroupViTVisionModel`].
    text_model_output (`BaseModelOutputWithPooling`):
        The output of the [`GroupViTTextModel`].
    vision_model_output (`BaseModelOutputWithPooling`):
        The output of the [`GroupViTVisionModel`].
    Nlosslogits_per_imagelogits_per_textsegmentation_logitstext_embedsimage_embedstext_model_outputvision_model_outputr   c                 H     t         fd j                         D              S )Nc              3   d   K   | ]'  }|d vr|   nt        |      j                          ) yw))r   r   N)getattrto_tuple).0kr   s     r#   	<genexpr>z/GroupViTModelOutput.to_tuple.<locals>.<genexpr>1  s=      
  LLDGRYZ^`aRbRkRkRmm
s   -0)tuplekeysr   s   `r#   r   zGroupViTModelOutput.to_tuple0  s#     
YY[
 
 	
r%   )r   r   r   __doc__r   r    FloatTensor__annotations__r   r   r   r   r   r   r   r   r   r   r    r%   r#   r   r     s    > &*D%

d
")15e''$.504OU&&-448**T18,0K""T)0-1L%##d*148186:3:
%* 
r%   r   c            	            e Zd ZdZ	 	 	 	 ddedeeeef   z  dedef fdZddej                  de	d	ej                  fd
Z
 xZS )GroupViTPatchEmbeddingsz#
    Image to Patch Embedding.
    
image_size
patch_sizenum_channels	embed_dimc                 ^   t         |           t        |t        j                  j
                        r|n||f}t        |t        j                  j
                        r|n||f}|d   |d   z  |d   |d   z  z  }|| _        || _        || _        t        j                  ||||      | _        y )Nr   r   )kernel_sizestride)rv   rw   r   r   r   r   r   r   num_patchesr   Conv2d
projection)r   r   r   r   r   r   r   s         r#   rw   z GroupViTPatchEmbeddings.__init__<  s     	#-j+//:R:R#SZZdfpYq
#-j+//:R:R#SZZdfpYq
!!}
15*Q-:VW=:XY$$&))L)\fgr%   pixel_valuesinterpolate_pos_encodingr   c                 8   |j                   \  }}}}|sV|| j                  d   k7  s|| j                  d   k7  r2t        d| d| d| j                  d    d| j                  d    d	      | j                  |      j	                  d      j                  dd      }|S )Nr   r   zInput image size (*z) doesn't match model ().rM   )rH   r   
ValueErrorr   flattenr   )r   r   r   r_   r   rZ   r[   r   s           r#   r   zGroupViTPatchEmbeddings.forwardM  s    2>2D2D/
L&%'++u8J/J (% 9+,Adooa.@-AE  OOL)11!4>>q!Dr%   )      r   i   F)r   r   r   r   rT   r   rw   r    Tensorboolr   r   r   s   @r#   r   r   7  sx     ,.hh %S/)h 	h
 h"	ELL 	D 	]b]i]i 	r%   r   c                        e Zd Zdef fdZdej                  dededej                  fdZddej                  d	e	dej                  fd
Z
 xZS )GroupViTVisionEmbeddingsrq   c                    t         |           t        |j                  |j                  |j
                  |j                        | _        | j                  j                  }t        j                  t        j                  d||j                              | _        t        j                  |j                        | _        t        j                   |j                  |j"                        | _        |j                  | _        || _        y )N)r   r   r   r   r   rt   )rv   rw   r   r   r   r   r{   patch_embeddingsr   r   	Parameterr    zerosposition_embeddingsDropoutdropoutrz   r|   	layernormrq   )r   rq   r   r   s      r#   rw   z!GroupViTVisionEmbeddings.__init__Z  s     7((((,,((	!
 ++77#%<<A{FL^L^0_#` zz&..1f&8&8f>S>ST ++r%   
embeddingsrZ   r[   r   c                 0   |j                   d   }| j                  j                   d   }t        j                  j	                         s||k(  r||k(  r| j                  S | j                  }|j                   d   }|| j
                  z  }|| j
                  z  }	t        |dz        }
|j                  d|
|
|      }|j                  dddd      }t        j                  j                  |||	fdd	      }|j                  dddd      j                  dd|      }|S )
a  
        This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution
        images. This method is also adapted to support torch.jit tracing and no class embeddings.

        Adapted from:
        - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and
        - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211
        r   rc   rN   r   r   rM   bicubicFrP   )rH   r   r    jit
is_tracingr   r   rW   re   r   r   rX   view)r   r   rZ   r[   r   num_positionspatch_pos_embedr,   
new_height	new_widthsqrt_num_positionss              r#   r   z1GroupViTVisionEmbeddings.interpolate_pos_encodingj  s#    !&&q)0066q9 yy##%+*F6UZ?+++22r"t.
T__,	&}c'9:)11!5GI[]`a)11!Q1=--33i(	 4 
 *11!Q1=BB1b#Nr%   r   r   c                 
   |j                   \  }}}}| j                  ||      }| j                  |      }|j                         \  }}}	|r|| j	                  |||      z   }n|| j
                  z   }| j                  |      }|S )N)r   )rH   r   r   rQ   r   r   r   )
r   r   r   r_   r   rZ   r[   r   seq_len_s
             r#   r   z GroupViTVisionEmbeddings.forward  s    2>2D2D/
L&%**<Rj*k
^^J/
!+!2
GQ $#d&C&CJPVX]&^^J#d&>&>>J\\*-
r%   r   )r   r   r   r   rw   r    r   rT   r   r   r   r   r   s   @r#   r   r   Y  sc    3  $5<< $ $UX $]b]i]i $LELL D ]b]i]i r%   r   c            	            e Zd Zdef fdZ	 	 	 d	dej                  dz  dej                  dz  dej                  dz  dej                  fdZ	 xZ
S )
GroupViTTextEmbeddingsrq   c                 N   t         |           |j                  }t        j                  |j
                  |      | _        t        j                  |j                  |      | _        | j                  dt        j                  |j                        j                  d      d       y )Nposition_idsr   rc   F)
persistent)rv   rw   r{   r   	Embedding
vocab_sizetoken_embeddingmax_position_embeddingsposition_embeddingregister_bufferr    r!   expandr   rq   r   r   s      r#   rw   zGroupViTTextEmbeddings.__init__  s    &&	!||F,=,=yI"$,,v/M/My"Y 	ELL)G)GHOOPWXej 	 	
r%   N	input_idsr  inputs_embedsr   c                 8   ||j                   d   n|j                   d   }| j                  j                  j                   d   }||kD  rt        d| d|       || j                  d d d |f   }|| j                  |      }| j                  |      }||z   }|S )Nrc   r   r   zRSequence length must be less than max_position_embeddings (got `sequence length`: z and max_position_embeddings: )rH   r  weightr   r  r  )r   r  r  r  
seq_lengthmax_position_embeddingr   r   s           r#   r   zGroupViTTextEmbeddings.forward  s     -6,AY__R(}GZGZ[]G^
!%!8!8!?!?!E!Ea!H..d,<=S<TV 
 ,,Q^<L  00;M"55lC"%88
r%   NNN)r   r   r   r   rw   r    
LongTensorr   r   r   r   r   s   @r#   r  r    sk    

1 

 .20426	##d* &&- ((4/	
 
r%   r  c            
           e Zd ZdZdededededef
 fdZed        Zd	 Z	dde
j                  de
j                  d
z  de
j                  fdZ	 	 dde
j                  de
j                  d
z  ded
z  dee
j                     fdZ xZS )GroupViTStagezMThis corresponds to the `GroupingLayer` class in the GroupViT implementation.rq   depthnum_prev_group_tokenr   r   c           	      f   t         |           || _        || _        |dkD  r:t	        j
                  t        j                  d||j                              | _	        nd | _	        t	        j                  t        |      D cg c]  }t        |       c}      | _        |dkD  rt        |||      | _        nd | _        |dkD  rc|dkD  r^t	        j                   t	        j"                  |j                  |j$                        t'        |||j                  dz  |            | _        y d | _        y c c}w )Nr   r   )rq   r   r   rt   rM   )rv   rw   r#  r   r   r   r    r   r{   group_token
ModuleListrangeGroupViTEncoderLayerlayersr   
downsample
Sequentialrz   r|   r   group_projector)r   rq   r#  r$  r   r   r
  r   s          r#   rw   zGroupViTStage.__init__  s    	
.Q!||EKK?FL^L^,_`D#Dmm5QV<$Xa%9&%A$XYQ1 /!1DO #DO!#!(;#%==V//V5J5JK )=v?Q?QUV?VXgh$D 
 $(D # %Ys    D.c                     | j                   d uS N)r&  r   s    r#   with_group_tokenzGroupViTStage.with_group_token  s    t++r%   c                 z    | j                   r,|d d d | j                   f   |d d | j                   d f   fS |d fS r/  )r0  r   )r   r   s     r#   split_xzGroupViTStage.split_x  sN      Q/4/////0!A8L8L7L7N4N2OOOd7Nr%   Nr   r&  r   c                 <    ||S t        j                  ||gd      S )Nr   r   )r    cat)r   r   r&  s      r#   concat_xzGroupViTStage.concat_x  s#    Hyy![)q11r%   hidden_statesprev_group_tokenoutput_attentionsc                    | j                   rM| j                  j                  |j                  d      dd      }| j                  || j	                  |      z   }nd}|}| j                  ||      }| j                  D ]  } ||d      }|d   } | j                  |      \  }}d}	| j                  | j                  ||      \  }}	||f}
|r|
|	fz   }
|
S )a  
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`): attention mask of size
                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
                `(config.encoder_attention_heads,)`.
            output_attentions (`bool`, *optional*):
                Whether or not to return the grouping tensors of Grouping block.
        r   rc   N)attention_mask)	r0  r&  r  rQ   r-  r5  r*  r2  r+  )r   r6  r7  r8  r&  r   cat_xlayer	layer_outr   outputss              r#   r   zGroupViTStage.forward   s       **11-2D2DQ2GRPK##/)D,@,@AQ,RRKa-[[ 	!EeD9IaLE	! e,;	??&??1k:LAyk",Gr%   r/  NF)r   r   r   r   r   rT   rw   propertyr0  r2  r    r   r5  r   r   r   r   r   r   s   @r#   r"  r"    s    W ($ (  ( "	 (
  (  (D , ,2%,, 2U\\D5H 2TYT`T` 2 15).	'||'  ,,-'  $;	'
 
u  	!'r%   r"  c            
            e Zd Z	 	 	 d
dededz  dedz  dedz  f fdZdej                  dej                  fd	Z xZ	S )r~   Nrq   r{   intermediate_sizeoutput_sizec                    t         |           || _        t        |j                     | _        ||n|j                  }||n|j                  }||n|}t        j                  ||      | _
        t        j                  ||      | _        y r/  )rv   rw   rq   r   
hidden_actactivation_fnr{   rB  r   r   fc1fc2)r   rq   r{   rB  rC  r   s        r#   rw   zGroupViTMLP.__init__+  s     	#F$5$56%0%<k&BTBT1B1N-TZTlTl%0%<k+99[*;<99.<r%   r6  r   c                 l    | j                  |      }| j                  |      }| j                  |      }|S r/  )rG  rF  rH  )r   r6  s     r#   r   zGroupViTMLP.forward;  s4    /**=9/r%   r  )
r   r   r   r   rT   rw   r    r   r   r   r   s   @r#   r~   r~   *  s`     #'(,"&=$= 4Z= :	=
 4Z= U\\ ell r%   r~   c                        e Zd Z fdZ xZS )r   c                 f    t         |   |j                  dd            }|j                  dd      S Nr   rM   )rv   r   r   )r   r   r   s     r#   r   zGroupViTMixerMLP.forwardC  s-    GOAKK1-.{{1a  r%   )r   r   r   r   r   r   s   @r#   r   r   B  s    ! !r%   r   c                       e Zd ZdZ fdZdej                  dedefdZ	 	 	 ddej                  d	ej                  dz  d
ej                  dz  de
dz  deej                  ej                  dz  eej                     dz  f   f
dZ xZS )rx   z=Multi-headed attention from 'Attention Is All You Need' paperc                 
   t         |           || _        |j                  | _        |j
                  | _        | j                  | j                  z  | _        | j                  | j                  z  | j                  k7  r&t        d| j                   d| j                   d      | j                  dz  | _	        |j                  | _        t        j                  | j                  | j                        | _        t        j                  | j                  | j                        | _        t        j                  | j                  | j                        | _        t        j                  | j                  | j                        | _        y )Nz;embed_dim must be divisible by num_heads (got `embed_dim`: z and `num_heads`: r   r   )rv   rw   rq   r{   r   num_attention_heads	num_headshead_dimr   r\   attention_dropoutr   r   r   r   r   r   out_projr   s     r#   rw   zGroupViTAttention.__init__K  s   ++33$..8==4>>)T^^;MdnnM] ^NN#2'  ]]D(
//ii?ii?ii?		$..$..Ar%   rF   r	  bszc                     |j                  ||| j                  | j                        j                  dd      j	                         S rL  )r  rP  rQ  r   rf   )r   rF   r	  rT  s       r#   _shapezGroupViTAttention._shape^  s7    {{3GQQRSUVWbbddr%   Nr6  r:  r   r8  r   c                 0   |j                         \  }}}|du}	| j                  |      | j                  z  }
|	rE| j                  | j	                  |      d|      }| j                  | j                  |      d|      }nD| j                  | j	                  |      d|      }| j                  | j                  |      d|      }|| j                  z  d| j                  f} | j                  |
||      j                  | }
 |j                  | } |j                  | }|j                  d      }t        j                  |
|j                  dd            }|j                         || j                  z  ||fk7  r/t        d|| j                  z  ||f d|j                                |{|j                         |d||fk7  r#t        d|d||f d|j                                |j                  || j                  ||      |z   }|j                  || j                  z  ||      }t        j                  j                  |d      }|r?|j                  || j                  ||      }|j                  || j                  z  ||      }nd}t        j                  j!                  || j                   | j"                  	      }t        j                  ||      }|j                         || j                  z  || j                  fk7  r7t        d
|| j                  || j                  f d|j                                |j                  || j                  || j                        }|j                  dd      }|j%                  |||      }| j'                  |      }||fS )z#Input shape: Batch x Time x ChannelNrc   r   rM   z$Attention weights should be of size z	, but is z!Attention mask should be of size r   )pr   z `attn_output` should be of size )rQ   r   r\   rV  r   r   rP  rQ  r  r    bmmr   r   r   r   r3   r   r   rW   rS  )r   r6  r:  r   r8  kwargsrT  tgt_lenr   is_cross_attentionquery_states
key_statesvalue_states
proj_shapesrc_lenattn_weightsattn_weights_reshaped
attn_probsattn_outputs                      r#   r   zGroupViTAttention.forwarda  s    #0"4"4"6Wi2$> {{=1DJJ>T[[1F%GSQJ;;t{{3H'I2sSLT[[%?SIJ;;t{{='A2sKLDNN*B>
Ct{{<#>CCZP$Z__j1
(|((*5//!$yyz/C/CAq/IJ3#7'"JJ6dnn8LgW^7_6` a %%'(* 
 %""$a'(BB 7a'8R7SS\]k]p]p]r\st  (,,S$..'7SVddL',,S4>>-A7GTL}},,\r,B
 %1$5$5c4>>7T[$\!055cDNN6JGU\]L$(!]]**<4<<RVR_R_*`
ii
L9#"6!OO2CRVR_R_3`2a b$$&') 
 "&&sDNNGT]]S!++Aq1!))#w	BmmK0111r%   )NNF)r   r   r   r   rw   r    r   rT   rV  r   r   r   r   r   r   s   @r#   rx   rx   H  s    GB&eU\\ eC ec e /3:>).H2||H2 t+H2  %0047	H2
  $;H2 
u||U\\D0%2E2LL	MH2r%   rx   c                        e Zd Zdef fdZ	 d
dej                  dej                  dedz  dee	   de
ej                     f
d	Z xZS )r)  rq   c                 D   t         |           |j                  | _        t	        |      | _        t        j                  | j                  |j                        | _	        t        |      | _        t        j                  | j                  |j                        | _        y rs   )rv   rw   r{   r   rx   	self_attnr   rz   r|   layer_norm1r~   r   layer_norm2r   s     r#   rw   zGroupViTEncoderLayer.__init__  sm    ++*62<<F<Q<QRv&<<F<Q<QRr%   r6  r:  r8  NrZ  r   c                     |}| j                  |      } | j                  d|||d|\  }}||z   }|}| j                  |      }| j                  |      }||z   }|f}|r||fz  }|S )aI  
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`): attention mask of size
                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
                `(config.encoder_attention_heads,)`.
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
        )r6  r:  r8  r   )ri  rh  rj  r   )r   r6  r:  r8  rZ  residualrb  r>  s           r#   r   zGroupViTEncoderLayer.forward  s    " !((7&4dnn '
')/'
 	'
#| !=0 ((7/ =0 "&Gr%   r   )r   r   r   r   rw   r    r   r   r   r   r   r   r   r   r   s   @r#   r)  r)    sh    S~ S */	&||& &  $;	&
 +,& 
u  	!&r%   r)  c                   R    e Zd ZU eed<   dZdZdZ ej                         d        Z
y)GroupViTPreTrainedModelrq   groupvit)imagetextTc                 T   | j                   j                  }t        |t        j                  t        j
                  f      rNt        j                  |j                  d|       |j                  t        j                  |j                         nt        |t        j                  t        j                  f      rt        j                  |j                         t        j                  |j                         t        |dd      ]t        j                  |j                         t        j                  |j                          t        j                  |j"                         | j                   j$                  }t        |t&              rt        j                  |j(                  j                  d|dz         t        j                  |j*                  j                  d|dz         t        j,                  |j.                  t1        j2                  |j.                  j4                  d         j7                  d             yt        |t8              r| j                   j$                  }|j:                  dz  d	|j                   j<                  z  dz  z  |z  }|j:                  dz  |z  }t        j                  |j>                  j                  |
       t        j                  |j@                  j                  |
       t        j                  |jB                  j                  |
       t        j                  |jD                  j                  |
       yt        |tF              r| j                   j$                  }|j                   jH                  dz  d	|j                   j<                  z  dz  z  |z  }d	|j                   jH                  z  dz  |z  }t        j                  |jJ                  j                  |
       t        j                  |jL                  j                  |
       yy)zInitialize the weightsrA   )meanstdNrunning_meang{Gz?rc   r  r   rM   )rt  )'rq   initializer_ranger   r   r   r   initnormal_r  biaszeros_rz   BatchNorm1dones_r   ru  running_varnum_batches_trackedinitializer_factorr  r  r  copy_r  r    r!   rH   r  rx   r   num_hidden_layersr   r   r   rS  r~   r{   rG  rH  )r   module
init_rangefactorin_proj_stdout_proj_stdfc_stds          r#   _init_weightsz%GroupViTPreTrainedModel._init_weights  s    [[22
fryy"))45LLSjA{{&FKK(r~~ >?KK$JJv}}%v~t4@F//0

6--.F667//f45LL//66SftmTLL2299RVWJJv**ELL9L9L9R9RSU9V,W,^,^_f,gh 12[[33F!++T1q6==;Z;Z7Z_c6cdgmmK",,d2f<LLL--;?LL--;?LL--;?LL//\B,[[33F!==44d:FMMDcDc@chl?lmpvvK&--333<vEFLL**7LL**< -r%   N)r   r   r   r   r   base_model_prefixinput_modalitiessupports_gradient_checkpointingr    rd   r  r   r%   r#   rn  rn    s4    "(&*#U]]_"= "=r%   rn  c                   x     e Zd Zdeddf fdZ	 	 	 d
dej                  dedz  dedz  dedz  dee	z  f
d	Z
 xZS )GroupViTVisionEncoderrq   r   Nc                 h   t         |           || _        t        j                  t        t        |j                              D cg c]P  }t        ||j                  |   |j                  |   |j                  |   |dkD  r|j                  |dz
     nd      R c}      | _        d| _        y c c}w )Nr   r   )rq   r#  r   r   r$  F)rv   rw   rq   r   r'  r(  r"   depthsr"  num_group_tokensnum_output_groupsstagesgradient_checkpointing)r   rq   ir   s      r#   rw   zGroupViTVisionEncoder.__init__  s    mm s6==12	  ! --*$*$;$;A$>%+%=%=a%@LMPQE)A)A!a%)HWX	
 ',#	s   AB/r6  output_hidden_statesr8  return_dictc                    ||n| j                   j                  }||n| j                   j                  }||n| j                   j                  }|rdnd }|rdnd }d }t	        | j
                        D ]3  \  }}	|r||fz   } |	|||      }
|
d   }|
d   }|s%|
d   +||
d   fz   }5 |r||fz   }|st        d |||fD              S t        |||      S )Nr   r   r   rM   c              3   &   K   | ]	  }||  y wr/  r   r   vs     r#   r   z0GroupViTVisionEncoder.forward.<locals>.<genexpr>@  s     gqYZYfg   last_hidden_stater6  rY   )rq   r8  r  use_return_dict	enumerater  r   r   )r   r6  r  r8  r  all_hidden_statesall_groupingsr   r  stagelayer_outputss              r#   r   zGroupViTVisionEncoder.forward  s     2C1N-TXT_T_TqTq$8$D $++JjJj 	 &1%<k$++B]B]"6BD/T!$++. 
	DHAu#$58H$H!!-?PQM)!,M(+L ]1%5%A -q1A0C C
	D   1]4D Dg]4E}$Uggg+;LYf
 	
r%   r  )r   r   r   r   rw   r    r   r   r   r   r   r   r   s   @r#   r  r    sl    ,3 , ,( -1)-#'%
||%
 #Tk%
  $;	%

 D[%
 
	 %
r%   r  c                        e Zd ZdZdef fdZ	 	 	 	 ddej                  dz  dedz  dedz  dedz  d	e	e
z  f
d
Z xZS )GroupViTTextEncoderz
    Transformer encoder consisting of `config.num_hidden_layers` self-attention layers. Each layer is a
    [`GroupViTEncoderLayer`].

    Args:
        config: GroupViTTextConfig
    rq   c                     t         |           || _        t        j                  t        |j                        D cg c]  }t        |       c}      | _        d| _	        y c c}w r?  )
rv   rw   rq   r   r'  r(  r  r)  r*  r  )r   rq   r
  r   s      r#   rw   zGroupViTTextEncoder.__init__O  sP    mm5QWQiQiKj$ka%9&%A$kl&+# %ls   A#Nr:  r8  r  r  r   c                    ||n| j                   j                  }||n| j                   j                  }||n| j                   j                  }|rdnd}|rdnd}|}	t	        | j
                        D ]*  \  }
}|r||	fz   } ||	|fd|i|}|d   }	|s"||d   fz   }, |r||	fz   }|st        d |	||fD              S t        |	||      S )ad  
        Args:
            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
                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.
            attention_mask (`torch.Tensor` 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)
            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.
        Nr   r8  r   r   c              3   &   K   | ]	  }||  y wr/  r   r  s     r#   r   z.GroupViTTextEncoder.forward.<locals>.<genexpr>  s     eqWXWder  r  )rq   r8  r  r  r  r*  r   r   )r   r  r:  r8  r  r  rZ  encoder_statesall_attentionsr6  idxencoder_layerr  s                r#   r   zGroupViTTextEncoder.forwardU  s   > 2C1N-TXT_T_TqTq$8$D $++JjJj 	 &1%<k$++B]B]30d%"+DKK"8 	FC#!/=2B!B) #4 	M *!,M !/=3C2E!E	F  +}.>>Ne]NN$Seee+>Vd
 	
r%   NNNN)r   r   r   r   r   rw   r    r   r   r   r   r   r   r   s   @r#   r  r  F  st    ,1 , /3)-,0#'?
 t+?
  $;	?

 #Tk?
 D[?
 
	 ?
r%   r  c                        e Zd Zdef fdZe	 	 	 	 	 	 ddej                  dz  dej                  dz  dej                  dz  dedz  dedz  d	edz  d
e	e
z  fd       Z xZS )GroupViTTextTransformerrq   c                 
   t         |   |       |j                  }t        |      | _        t        |      | _        t        j                  ||j                        | _
        |j                  | _        | j                          y rs   )rv   rw   r{   r  r   r  encoderr   rz   r|   final_layer_normeos_token_id	post_initr  s      r#   rw   z GroupViTTextTransformer.__init__  sf     &&	08*62 "YF<Q<Q R #//r%   Nr  r:  r  r8  r  r  r   c           
         ||n| j                   j                  }||n| j                   j                  }||n| j                   j                  }|t	        d      |j                         }|j                  d|d         }| j                  ||      }	t        | j                   |	|t        j                  |	j                  d   |	j                        d       }|j                  dd         | j                  d|	||||dd	|}
|
d
   }| j                  |      }| j                   dk(  rm|t        j                  |j                  d
   |j                        |j#                  t        j$                  |j                        j'                  d      f   }n|t        j                  |j                  d
   |j                        |j#                  t        j$                  |j                        | j                   k(  j%                         j'                  d      f   }|s
||f|
dd  z   S t)        |||
j*                  |
j,                        S )NzYou have to specify input_idsrc   )r  r  r   r   )rq   r  r:  cache_positionpast_key_values	is_causalT)r  r:  r8  r  r  r  r   rM   )rB   r   r   r  pooler_outputr6  rY   r   )rq   r8  r  r  r   rQ   r  r   r	   r    r!   rH   r   popr  r  r  torT   argmaxr   r6  rY   )r   r  r:  r  r8  r  r  rZ  input_shaper6  encoder_outputsr  pooled_outputs                r#   r   zGroupViTTextTransformer.forward  sH    2C1N-TXT_T_TqTq$8$D $++JjJj 	 &1%<k$++B]B]<==nn&NN2{27	),W+;;') <<(;(;A(>}G[G[\ 
 	

;%&$,, 
')/!5#
 
 ,A. 112CD! ..44Q7@Q@X@XY5995F5M5MNUUZ\U]_M ..44Q7@Q@X@XY EII6G6N6NOSWSdSddB!M %}58KKK)/')77&11	
 	
r%   NNNNNN)r   r   r   r   rw   r   r    r   r   r   r   r   r   r   s   @r#   r  r    s    
1 
  *..2,0)-,0#'L
<<$&L
 t+L
 llT)	L

  $;L
 #TkL
 D[L
 
+	+L
 L
r%   r  c                        e Zd ZU eed<   dZdef fdZdej                  fdZ	d Z
e	 	 	 	 	 	 ddej                  dz  d	ej                  dz  d
ej                  dz  dedz  dedz  dedz  deez  fd       Z xZS )GroupViTTextModelrq   )rq  c                 d    t         |   |       t        |      | _        | j	                          y r/  )rv   rw   r  
text_modelr  r   s     r#   rw   zGroupViTTextModel.__init__  s&     1&9r%   r   c                 B    | j                   j                  j                  S r/  r  r   r  r   s    r#   get_input_embeddingsz&GroupViTTextModel.get_input_embeddings  s    ))999r%   c                 :    || j                   j                  _        y r/  r  )r   r   s     r#   set_input_embeddingsz&GroupViTTextModel.set_input_embeddings  s    5:""2r%   Nr  r:  r  r8  r  r  c                 0    | j                  ||||||      S )a9  
        Examples:

        ```python
        >>> from transformers import CLIPTokenizer, GroupViTTextModel

        >>> tokenizer = CLIPTokenizer.from_pretrained("nvidia/groupvit-gcc-yfcc")
        >>> model = GroupViTTextModel.from_pretrained("nvidia/groupvit-gcc-yfcc")

        >>> inputs = tokenizer(["a photo of a cat", "a photo of a dog"], padding=True, return_tensors="pt")

        >>> outputs = model(**inputs)
        >>> last_hidden_state = outputs.last_hidden_state
        >>> pooled_output = outputs.pooler_output  # pooled (EOS token) states
        ```r  r:  r  r8  r  r  )r  )r   r  r:  r  r8  r  r  rZ  s           r#   r   zGroupViTTextModel.forward  s,    4 )%/!5#  
 	
r%   r  )r   r   r   r   r   r  rw   r   Moduler  r  r   r    r   r   r   r   r   r   r   s   @r#   r  r    s     1 :bii :;  *..2,0)-,0#' 
<<$& 
 t+ 
 llT)	 

  $; 
 #Tk 
 D[ 
 
+	+ 
  
r%   r  c                        e Zd Zdef fdZe	 	 	 	 d
dej                  dz  dedz  dedz  dedz  de	e
z  f
d	       Z xZS )GroupViTVisionTransformerrq   c                     t         |           || _        |j                  }t	        |      | _        t        |      | _        t        j                  ||j                        | _        y rs   )rv   rw   rq   r{   r   r   r  r  r   rz   r|   r   r  s      r#   rw   z"GroupViTVisionTransformer.__init__)  sP    &&	26:,V4iV5J5JKr%   Nr   r  r8  r  r   c                    ||n| j                   j                  }||n| j                   j                  }||n| j                   j                  }|t	        d      | j                  |      }| j                  ||||      }|d   }| j                  |      }|j                  d      }|s
||f|dd  z   S t        |||j                  |j                        S )Nz You have to specify pixel_values)r6  r  r8  r  r   r   r   r  )rq   r8  r  r  r   r   r  r   rs  r   r6  rY   )	r   r   r  r8  r  r6  r  r  r  s	            r#   r   z!GroupViTVisionTransformer.forward2  s    2C1N-TXT_T_TqTq$8$D $++JjJj 	 &1%<k$++B]B]?@@5,,'!5/#	 ' 
 ,A. !NN+<=)..1.5%}58KKK)/')77&11	
 	
r%   r  )r   r   r   r   rw   r   r    r   r   r   r   r   r   r   s   @r#   r  r  (  s    L3 L  26,0)-#''
''$.'
 #Tk'
  $;	'

 D['
 
+	+'
 '
r%   r  c                        e Zd ZU eed<   dZdZdef fdZdefdZ	e
	 	 	 	 ddej                  dz  dedz  d	edz  d
edz  deez  f
d       Z xZS )GroupViTVisionModelrq   r   )rp  c                 d    t         |   |       t        |      | _        | j	                          y r/  )rv   rw   r  vision_modelr  r   s     r#   rw   zGroupViTVisionModel.__init__b  s'     5f=r%   r   c                 B    | j                   j                  j                  S r/  )r  r   r   r   s    r#   r  z(GroupViTVisionModel.get_input_embeddingsh  s      ++<<<r%   Nr8  r  r  c                 ,    | j                  ||||      S )a)  
        Examples:

        ```python
        >>> from PIL import Image
        >>> import httpx
        >>> from io import BytesIO
        >>> from transformers import AutoProcessor, GroupViTVisionModel

        >>> processor = AutoProcessor.from_pretrained("nvidia/groupvit-gcc-yfcc")
        >>> model = GroupViTVisionModel.from_pretrained("nvidia/groupvit-gcc-yfcc")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> with httpx.stream("GET", url) as response:
        ...     image = Image.open(BytesIO(response.read()))

        >>> inputs = processor(images=image, return_tensors="pt")

        >>> outputs = model(**inputs)
        >>> last_hidden_state = outputs.last_hidden_state
        >>> pooled_output = outputs.pooler_output  # pooled CLS states
        ```r   r8  r  r  )r  )r   r   r8  r  r  rZ  s         r#   r   zGroupViTVisionModel.forwardk  s(    >   %/!5#	 ! 
 	
r%   r  )r   r   r   r   r   main_input_namer  rw   r   r  r   r    r   r   r   r   r   r   r   s   @r#   r  r  ]  s      $O!3 =&= =  26)-,0#'#
''$.#
  $;#
 #Tk	#

 D[#
 
+	+#
 #
r%   r  c                       e Zd ZU eed<   def fdZee	 	 ddej                  dej                  dz  dej                  dz  de
e   deez  f
d	              Zeed
ej                  de
e   deez  fd              Ze	 	 	 	 	 	 	 	 	 ddej                   dz  d
ej"                  dz  dej                  dz  dej                   dz  dedz  dedz  dedz  dedz  dedz  deez  fd       Z xZS )GroupViTModelrq   c           
      6   t         |   |       t        |j                  t              s"t        dt        |j                         d      t        |j                  t              s"t        dt        |j                         d      |j                  }|j                  }|j                  | _	        |j                  | _
        |j                  | _        |j                  | _        t        |      | _        t!        |      | _        t%        j&                  t%        j(                  | j                  | j                  d      t%        j*                  | j                        t%        j,                  d      t%        j(                  | j                  | j                  d            | _        t%        j&                  t%        j(                  | j                  | j                  d      t%        j*                  | j                        t%        j,                  d      t%        j(                  | j                  | j                  d            | _        t%        j2                  t5        j6                  | j8                  j:                              | _        | j?                          y )NzOconfig.text_config is expected to be of type GroupViTTextConfig but is of type .zSconfig.vision_config is expected to be of type GroupViTVisionConfig but is of type T)ry  )inplace) rv   rw   r   text_configr   	TypeErrortypevision_configr   projection_dimprojection_intermediate_dimr{   text_embed_dimvision_embed_dimr  r  r  r  r   r,  r   r{  ReLUvisual_projectiontext_projectionr   r    rF   rq   logit_scale_init_valuelogit_scaler  )r   rq   r  r  r   s       r#   rw   zGroupViTModel.__init__  s    &,,.@A++,-Q0 
 &..0DE--./q2 
 ((,,$33+1+M+M()55 - 9 91+>5mD!#IId++T-M-MTXYNN4;;<GGD!IId668K8KRVW	"
  "}}IId))4+K+KRVWNN4;;<GGD!IId668K8KRVW	 
 <<T[[5W5W(XY 	r%   Nr  r:  r  rZ  r   c                 x     | j                   d|||dd|}|j                  }| j                  |      |_        |S )a  
        Examples:

        ```python
        >>> import torch
        >>> from transformers import CLIPTokenizer, GroupViTModel

        >>> model = GroupViTModel.from_pretrained("nvidia/groupvit-gcc-yfcc")
        >>> tokenizer = CLIPTokenizer.from_pretrained("nvidia/groupvit-gcc-yfcc")

        >>> inputs = tokenizer(["a photo of a cat", "a photo of a dog"], padding=True, return_tensors="pt")
        >>> with torch.inference_mode():
        ...     text_features = model.get_text_features(**inputs)
        ```T)r  r:  r  r  r   )r  r  r  )r   r  r:  r  rZ  text_outputsr  s          r#   get_text_featureszGroupViTModel.get_text_features  sV    . 4C4?? 4
)%	4

 4
 %22%)%9%9-%H"r%   r   c                 p     | j                   |fddi|}| j                  |j                        |_        |S )a  
        Examples:

        ```python
        >>> import torch
        >>> from transformers import AutoProcessor, GroupViTModel
        >>> from transformers.image_utils import load_image

        >>> model = GroupViTModel.from_pretrained("nvidia/groupvit-gcc-yfcc")
        >>> processor = AutoProcessor.from_pretrained("nvidia/groupvit-gcc-yfcc")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = load_image(url)

        >>> inputs = processor(images=image, return_tensors="pt")

        >>> with torch.inference_mode():
        ...     image_features = model.get_image_features(**inputs)
        ```r  T)r  r  r  )r   r   rZ  vision_outputss       r#   get_image_featuresz GroupViTModel.get_image_features  sB    4 6GT5F5F|5pae5pio5p'+'='=n>Z>Z'[$r%   return_lossr8  r  output_segmentationr  c
           
         ||n| j                   j                  }||n| j                   j                  }|rd}||n| j                   j                  }|	|	n| j                   j                  }	| j                  ||||	      }| j                  ||||||	      }|d   }| j                  |      }|d   }| j                  |      }||j                  dd      z  }||j                  dd      z  }| j                  j                         }t        j                  ||j                               |z  }|j                         }d}|rb|d   }| j                  |j                  d|j                   d               }|r|d	   }n|d
   }t#        ||j                   d
d       }||j                  dd      z  }t        j                  ||j                               |z  }|j                  |j                   d   d|j                   d         j%                  dd
d      }|j                  |j                   d   |j                   d   d      }t        j                  ||      |z  }|j                  |j                   d   |j                   d   |j                   d
   |j                   d	         }d}|rt'        |      }|	s|
|||||||f}n||||||f}||f|z   S |S t)        ||||||||      S )a  
        return_loss (`bool`, *optional*):
            Whether or not to return the contrastive loss.
        output_segmentation (`bool`, *optional*):
            Whether or not to return the segmentation logits.

        Examples:

        ```python
        >>> from PIL import Image
        >>> import httpx
        >>> from io import BytesIO
        >>> from transformers import AutoProcessor, GroupViTModel

        >>> model = GroupViTModel.from_pretrained("nvidia/groupvit-gcc-yfcc")
        >>> processor = AutoProcessor.from_pretrained("nvidia/groupvit-gcc-yfcc")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> with httpx.stream("GET", url) as response:
        ...     image = Image.open(BytesIO(response.read()))

        >>> inputs = processor(
        ...     text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True
        ... )

        >>> outputs = model(**inputs)
        >>> logits_per_image = outputs.logits_per_image  # this is the image-text similarity score
        >>> probs = logits_per_image.softmax(dim=1)  # we can take the softmax to get the label probabilities
        ```NTr  r  r   rc   r   r   r   rM   )r   r   r   r   r   r   r   r   )rq   r8  r  r  r  r  r  r  r  normr  expr    matmulr(   rW   rH   rn   re   r+   r   )r   r  r   r:  r  r  r8  r  r  r  rZ  r  r  r   r   r  r   r   
seg_logitsimage_group_embedsrY   groupinglogits_per_image_groupflatten_groupingr   outputs                             r#   r   zGroupViTModel.forward  sn   X 2C1N-TXT_T_TqTq#6#BHgHg 	  $$8$D $++JjJj 	 &1%<k$++B]B]**%/!5#	 + 
 )%/!5# ' 
 &a(--l;"1o**;7 $l&7&7B&7&MM!K$4$4T$4$JJ &&**,,,{LNN4DES*,,.
 "0!2!%!7!78J8R8RSUWiWoWoprWs8t!u#+A.
+A.
3J@R@RSTSU@VWH "46H6M6MRT^b6M6c!c%*\\2Dkmmo%VYd%d"%;%C%C""1%r;+<+<Q+?&gaA #
  (//q0A8>>RSCTVXY &<>NOR]]J#++  #Z%5%5a%8(..:KX^^\]M^J  1D%$#  " +O[,Xdftu)-)9TGf$EvE"-+ *#%* .	
 		
r%   )NN)	NNNNNNNNN)r   r   r   r   r   rw   r   r   r    r   r   r   r   r   r  r  r   r   r   r   r   r   r   s   @r#   r  r    s   )~ )V  /3,0	<< t+ llT)	
 +, 
+	+  B ll +, 
+	+	  :  .215.204#')-,0+/#'Q
##d*Q
 ''$.Q
 t+	Q

 &&-Q
 D[Q
  $;Q
 #TkQ
 "D[Q
 D[Q
 
$	$Q
 Q
r%   r  )r  rn  r  r  )r   Frc   r   )Ir   collections.abcr   dataclassesr   typingr   numpyrU   r    r    r   rw  activationsr   masking_utilsr	   modeling_layersr
   modeling_outputsr   r   modeling_utilsr   processing_utilsr   utilsr   r   r   r   r   r   configuration_groupvitr   r   r   
get_loggerr   loggerr   r$   r+   rT   r=   floatr   rK   ra   rn   r  rp   r   r   r   r   r   r  r"  r~   r   rx   r)  rn  r  r  r  r  r  r  r  __all__r   r%   r#   <module>r     s_     !     & ! / 9 K - & j j \ \ 
		H	%
`U\\ `ell `
-ell -u|| - C 5<< e t RU _d_k_k ,<:"))  -bii -`4+")) 4+n -
+ -
  -
`bii DGryy GV%RYY %P[BII [|")) 0!{ !a2		 a2J/5 /d )=o )= )=X7
BII 7
tN
")) N
bZ
5 Z
z1
/ 1
h2
		 2
j2
1 2
j B
+ B
 B
J cr%   