
    3jC                        S r SSKrSSKJr  SSKJr  SSKJr  SSKJ	r	J
r
  SSKJrJrJrJr  SS	KJr  SS
KJr  SSKJrJrJr  SSKJrJr  SSKJr  SSKJr  \R>                  " \ 5      r! " S S\RD                  5      r# " S S\RH                  5      r% " S S\RD                  5      r& " S S\RD                  5      r' " S S\RD                  5      r( " S S\RD                  5      r)\ " S S\5      5       r* " S S\*5      r+\ " S  S!\*5      5       r,\" S"S#9 " S$ S%\*5      5       r-\" S&S#9 " S' S(\	\*5      5       r./ S)Qr/g)*zPyTorch ConvNextV2 model.    N)nn   )initialization)ACT2FN)BackboneMixinfilter_output_hidden_states)BackboneOutputBaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)Unpack)TransformersKwargsauto_docstringlogging)can_return_tuplemerge_with_config_defaults)capture_outputs   )ConvNextV2Configc                   n   ^  \ rS rSrSrS\4U 4S jjrS\R                  S\R                  4S jr	Sr
U =r$ )	ConvNextV2GRN'   z)GRN (Global Response Normalization) layerdimc                    > [         TU ]  5         [        R                  " [        R
                  " SSSU5      5      U l        [        R                  " [        R
                  " SSSU5      5      U l        g )Nr   )super__init__r   	Parametertorchzerosweightbias)selfr   	__class__s     l/home/wildlama/miniconda3/lib/python3.13/site-packages/transformers/models/convnextv2/modeling_convnextv2.pyr   ConvNextV2GRN.__init__*   sL    ll5;;q!Q#<=LLQ1c!:;	    hidden_statesreturnc                     [         R                  R                  USSSS9nX"R                  SSS9S-   -  nU R                  X-  -  U R
                  -   U-   nU$ )N   )r   r+   T)ordr   keepdim)r   r-   ư>)r   linalgvector_normmeanr!   r"   )r#   r(   global_featuresnorm_featuress       r%   forwardConvNextV2GRN.forward/   se    ,,22=aV]a2b'+?+?BPT+?+UX\+\]}'DE		QTaar'   )r"   r!   )__name__
__module____qualname____firstlineno____doc__intr   r   FloatTensorr5   __static_attributes____classcell__r$   s   @r%   r   r   '   s6    3<C <
U%6%6 5;L;L  r'   r   c                   v   ^  \ rS rSrSrSSS.U 4S jjrS\R                  S\R                  4U 4S	 jjrS
r	U =r
$ )ConvNextV2LayerNorm9   a5  LayerNorm that supports two data formats: channels_last (default) or channels_first.
The ordering of the dimensions in the inputs. channels_last corresponds to inputs with shape (batch_size, height,
width, channels) while channels_first corresponds to inputs with shape (batch_size, channels, height, width).
r/   channels_lastepsdata_formatc                `   > [         TU ]  " U4SU0UD6  US;  a  [        SU 35      eX0l        g )NrF   )rD   channels_firstzUnsupported data format: )r   r   NotImplementedErrorrG   )r#   normalized_shaperF   rG   kwargsr$   s        r%   r   ConvNextV2LayerNorm.__init__?   s=    )=s=f=AA%(A+&OPP&r'   featuresr)   c                    > U R                   S:X  a9  UR                  SSSS5      n[        TU ]  U5      nUR                  SSSS5      nU$ [        TU ]  U5      nU$ )zt
Args:
    features: Tensor of shape (batch_size, channels, height, width) OR (batch_size, height, width, channels)
rI   r   r+   r   r   )rG   permuter   r5   )r#   rN   r$   s     r%   r5   ConvNextV2LayerNorm.forwardE   sj    
 //''1a3Hwx0H''1a3H  wx0Hr'   rG   r7   r8   r9   r:   r;   r   r   Tensorr5   r>   r?   r@   s   @r%   rB   rB   9   s9    
 15/ ' '   r'   rB   c                   f   ^  \ rS rSrSrU 4S jrS\R                  S\R                  4S jr	Sr
U =r$ )ConvNextV2EmbeddingsT   zThis class is comparable to (and inspired by) the SwinEmbeddings class
found in src/transformers/models/swin/modeling_swin.py.
c                   > [         TU ]  5         [        R                  " UR                  UR
                  S   UR                  UR                  S9U l        [        UR
                  S   SSS9U l	        UR                  U l        g )Nr   kernel_sizestrider/   rI   rE   )
r   r   r   Conv2dnum_channelshidden_sizes
patch_sizepatch_embeddingsrB   	layernormr#   configr$   s     r%   r   ConvNextV2Embeddings.__init__Y   sr     "		!4!4Q!7VEVEV_e_p_p!
 -V-@-@-C[kl"//r'   pixel_valuesr)   c                     UR                   S   nX R                  :w  a  [        S5      eU R                  U5      nU R	                  U5      nU$ )Nr   zeMake sure that the channel dimension of the pixel values match with the one set in the configuration.)shaper]   
ValueErrorr`   ra   )r#   re   r]   
embeddingss       r%   r5   ConvNextV2Embeddings.forwarda   sT    #))!,,,,w  **<8
^^J/
r'   )ra   r]   r`   )r7   r8   r9   r:   r;   r   r   r=   rT   r5   r>   r?   r@   s   @r%   rV   rV   T   s/    0E$5$5 %,,  r'   rV   c                      ^  \ rS rSrSrSS\SS4U 4S jjjrS\R                  S\R                  4S jr	S\
4S	 jrS
rU =r$ )ConvNextV2DropPathm   zStochastic depth (DropPath) per sample, for residual blocks.

Identity when ``drop_prob`` is 0 or outside training. See `Deep Networks with Stochastic Depth
<https://arxiv.org/abs/1603.09382>`_.
	drop_probr)   Nc                 .   > [         TU ]  5         Xl        g N)r   r   rn   )r#   rn   r$   s     r%   r   ConvNextV2DropPath.__init__t   s    "r'   r(   c                 V   U R                   S:X  d  U R                  (       d  U$ SU R                   -
  nUR                  S   4SUR                  S-
  -  -   n[        R
                  " X1R                  UR                  S9n[        R                  " XB-   5      nUR                  U5      U-  $ )N        r   r   )r   )dtypedevice)
rn   trainingrg   ndimr   randrt   ru   floordiv)r#   r(   	keep_probrg   random_tensors        r%   r5   ConvNextV2DropPath.forwardx   s    >>S   &	$$Q')DM4F4F4J,KK

50C0CML`L`aM$=>  +m;;r'   c                      SU R                    3$ )Nzp=rn   )r#   s    r%   
extra_reprConvNextV2DropPath.extra_repr   s    DNN#$$r'   r   )rs   )r7   r8   r9   r:   r;   floatr   r   rT   r5   strr   r>   r?   r@   s   @r%   rl   rl   m   sL    #% #$ # #<U\\ <ell <%C % %r'   rl   c                   j   ^  \ rS rSrSrSU 4S jjrS\R                  S\R                  4S jrSr	U =r
$ )	ConvNextV2Layer   a  This corresponds to the `Block` class in the original implementation.

There are two equivalent implementations: [DwConv, LayerNorm (channels_first), Conv, GELU,1x1 Conv]; all in (N, C,
H, W) (2) [DwConv, Permute to (N, H, W, C), LayerNorm (channels_last), Linear, GELU, Linear]; Permute back

The authors used (2) as they find it slightly faster in PyTorch.

Args:
    config ([`ConvNextV2Config`]): Model configuration class.
    dim (`int`): Number of input channels.
    drop_path (`float`): Stochastic depth rate. Default: 0.0.
c                   > [         TU ]  5         [        R                  " X"SSUS9U l        [        USS9U l        [        R                  " USU-  5      U l        [        UR                     U l        [        SU-  5      U l        [        R                  " SU-  U5      U l        US:  a  [        U5      U l        g [        R                   " 5       U l        g )N   r   )rZ   paddinggroupsr/   rF      rs   )r   r   r   r\   dwconvrB   ra   Linearpwconv1r   
hidden_actactr   grnpwconv2rl   Identity	drop_path)r#   rc   r   r   r$   s       r%   r   ConvNextV2Layer.__init__   s    iia3O,Sd;yya#g.&++, S)yyS#.:Cc/+I6r{{}r'   rN   r)   c                 L   UnU R                  U5      nUR                  SSSS5      nU R                  U5      nU R                  U5      nU R	                  U5      nU R                  U5      nU R                  U5      nUR                  SSSS5      nX R                  U5      -   nU$ )Nr   r+   r   r   )r   rP   ra   r   r   r   r   r   )r#   rN   residuals      r%   r5   ConvNextV2Layer.forward   s    ;;x(##Aq!Q/>>(+<<)88H%88H%<<)##Aq!Q/nnX66r'   )r   r   r   r   ra   r   r   )r   rS   r@   s   @r%   r   r      s.    
]   r'   r   c                   j   ^  \ rS rSrSrSU 4S jjrS\R                  S\R                  4S jrSr	U =r
$ )	ConvNextV2Stage   a  ConvNeXTV2 stage, consisting of an optional downsampling layer + multiple residual blocks.

Args:
    config ([`ConvNextV2Config`]): Model configuration class.
    in_channels (`int`): Number of input channels.
    out_channels (`int`): Number of output channels.
    depth (`int`): Number of residual blocks.
    drop_path_rates(`list[float]`): Stochastic depth rates for each layer.
c                   > [         T	U ]  5         X#:w  d  US:  a:  [        R                  " [	        USSS9[        R
                  " X#XES9/5      U l        O[        R                  " 5       U l        U=(       d    S/U-  n[        R                  " [        U5       Vs/ s H  n[        XXx   S9PM     sn5      U l	        g s  snf )Nr   r/   rI   rE   rY   rs   )r   r   )
r   r   r   
ModuleListrB   r\   downsampling_layerranger   layers)
r#   rc   in_channelsout_channelsrZ   r[   depthdrop_path_ratesjr$   s
            r%   r   ConvNextV2Stage.__init__   s    &&1*&(mm'K[\IIk[`'D# ')mmoD#):cUU]mm^cdi^jk^jYZ_VAST^jk
ks   B>rN   r)   c                 r    U R                    H  nU" U5      nM     U R                   H  nU" U5      nM     U$ rp   r   r   )r#   rN   layers      r%   r5   ConvNextV2Stage.forward   s7    ,,EXH -[[EXH !r'   r   )r+   r+   r+   NrS   r@   s   @r%   r   r      s-    
"   r'   r   c                   n   ^  \ rS rSr% \\S'   SrSrSrS/r	\
R                  " 5       U 4S j5       rSrU =r$ )	ConvNextV2PreTrainedModel   rc   
convnextv2re   )imager   c                    > [         TU ]  U5        [        U[        5      (       aA  [        R
                  " UR                  5        [        R
                  " UR                  5        gg)zInitialize the weightsN)r   _init_weights
isinstancer   initzeros_r!   r"   )r#   moduler$   s     r%   r   'ConvNextV2PreTrainedModel._init_weights   sD     	f%fm,,KK&KK$ -r'    )r7   r8   r9   r:   r   __annotations__base_model_prefixmain_input_nameinput_modalities_no_split_modulesr   no_gradr   r>   r?   r@   s   @r%   r   r      s;    $$O!*+
]]_% %r'   r   c                      ^  \ rS rSrSrS\0rU 4S jr\\	" SS9S\
R                  S\\   S\4S j5       5       rS	rU =r$ )
ConvNextV2Encoder   r(   c           
      P  > [         TU ]  U5        [        R                  " 5       U l        [
        R                  " SUR                  [        UR                  5      SS9R                  UR                  5       Vs/ s H  nUR                  5       PM     nnUR                  S   n[        UR                  5       HT  nUR                  U   n[        UUUUS:  a  SOSUR                  U   X5   S9nU R                  R!                  U5        UnMV     U R#                  5         g s  snf )Nr   cpu)ru   r+   r   )r   r   r[   r   r   )r   r   r   r   stagesr   linspacedrop_path_ratesumdepthssplittolistr^   r   
num_stagesr   append	post_init)	r#   rc   xr   prev_chsiout_chsstager$   s	           r%   r   ConvNextV2Encoder.__init__   s    mmo ^^Av'<'<c&-->PY^_eeflfsfst
t HHJt 	 
 &&q)v(()A))!,G#$$EqqmmA& / 2E KKu%H * 	%
s   :D#F)tie_last_hidden_statesrL   r)   c                 J    U R                    H  nU" U5      nM     [        US9$ )N)last_hidden_state)r   r
   )r#   r(   rL   layer_modules       r%   r5   ConvNextV2Encoder.forward  s)     !KKL(7M ( .NNr'   )r   )r7   r8   r9   r:   r   r   _can_record_outputsr   r   r   r   rT   r   r   r
   r5   r>   r?   r@   s   @r%   r   r      sd    %O*O<.  E2O||O +,O 
(	O 3  Or'   r   c            	       x   ^  \ rS rSrU 4S jr\\ S	S\R                  S-  S\	\
   S\4S jj5       5       rSrU =r$ )
ConvNextV2Modeli  c                    > [         TU ]  U5        Xl        [        U5      U l        [        U5      U l        [        R                  " UR                  S   UR                  S9U l        U R                  5         g )Nr.   r   )r   r   rc   rV   ri   r   encoderr   	LayerNormr^   layer_norm_epsra   r   rb   s     r%   r   ConvNextV2Model.__init__  s^     .v6(0 f&9&9"&=6CXCXY 	r'   Nre   rL   r)   c                     Uc  [        S5      eU R                  U5      nU R                  " U40 UD6nUR                  nU R	                  UR                  SS/5      5      n[        UUUR                  S9$ )Nz You have to specify pixel_valuesr.   )r   pooler_outputr(   )rh   ri   r   r   ra   r2   r   r(   )r#   re   rL   embedding_outputencoder_outputsr   pooled_outputs          r%   r5   ConvNextV2Model.forward  s    
 ?@@??<8:>,,GW:b[a:b+== '8'='=r2h'GH7/')77
 	
r'   )rc   ri   r   ra   rp   )r7   r8   r9   r:   r   r   r   r   r=   r   r   r   r5   r>   r?   r@   s   @r%   r   r     sP     7;
!--4
GMN`Ga
	1
  
r'   r   z
    ConvNextV2 Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for
    ImageNet.
    )custom_introc            	          ^  \ rS rSrSrU 4S jr\\ S
S\R                  S-  S\R                  S-  S\4S jj5       5       rS	rU =r$ ) ConvNextV2ForImageClassificationi5  Fc                 B  > [         TU ]  U5        UR                  U l        [        U5      U l        UR                  S:  a4  [
        R                  " UR                  S   UR                  5      U l        O[
        R                  " 5       U l        U R                  5         g )Nr   r.   )r   r   
num_labelsr   r   r   r   r^   
classifierr   r   rb   s     r%   r   )ConvNextV2ForImageClassification.__init__?  su      ++)&1 q  ii(;(;B(?ARARSDO kkmDO 	r'   Nre   labelsr)   c                     U R                   " U40 UD6nUR                  nU R                  U5      nSnUb  U R                  X&U R                  S9n[        UUUR                  S9$ )ab  
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
    Labels for computing the image classification/regression loss. Indices should be in `[0, ...,
    config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
    `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
N)r   pooled_logitsrc   )losslogitsr(   )r   r   r   loss_functionrc   r   r(   )r#   re   r   rL   outputsr   r   r   s           r%   r5   (ConvNextV2ForImageClassification.forwardN  su     =AOOL<c\b<c--/%%VRVR]R]%^D3!//
 	
r'   )r   r   r   )NN)r7   r8   r9   r:   accepts_loss_kwargsr   r   r   r   r=   
LongTensorr   r5   r>   r?   r@   s   @r%   r   r   5  s^       _c
!--4
EJEUEUX\E\
	-
  
r'   r   zT
    ConvNeXT V2 backbone, to be used with frameworks like DETR and MaskFormer.
    c            	       z   ^  \ rS rSrSrU 4S jr\\\S\	R                  S\\   S\4S j5       5       5       rSrU =r$ )	ConvNextV2Backboneih  Fc                 l  > [         TU ]  U5        [        U5      U l        [	        U5      U l        UR                  S   /UR                  -   U l        0 n[        U R                  U R                  5       H  u  p4[        USS9X#'   M     [        R                  " U5      U l        U R                  5         g )Nr   rI   rR   )r   r   rV   ri   r   r   r^   num_featureszipout_featureschannelsrB   r   
ModuleDicthidden_states_normsr   )r#   rc   r  r   r]   r$   s        r%   r   ConvNextV2Backbone.__init__q  s     .v6(0#0034v7J7JJ !#&t'8'8$--#HE)<\Wg)h& $I#%==1D#E  	r'   re   rL   r)   c                 8   U R                  U5      nU R                  " U40 UD6nUR                  n/ n[        U R                  U5       H<  u  pxXpR
                  ;   d  M  U R                  U   " U5      nUR                  U5        M>     [        [        U5      US9$ )aK  
Examples:

```python
>>> from transformers import AutoImageProcessor, AutoBackbone
>>> import torch
>>> from PIL import Image
>>> import httpx
>>> from io import BytesIO

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

>>> processor = AutoImageProcessor.from_pretrained("facebook/convnextv2-tiny-1k-224")
>>> model = AutoBackbone.from_pretrained("facebook/convnextv2-tiny-1k-224")

>>> inputs = processor(image, return_tensors="pt")
>>> outputs = model(**inputs)
```)feature_mapsr(   )
ri   r   r(   r   stage_namesr   r  r   r	   tuple)	r#   re   rL   r   r   r(   r  r   hidden_states	            r%   r5   ConvNextV2Backbone.forward  s    8  ??<8:>,,GW:b[a:b'55#&t'7'7#GE)))#77>|L##L1 $H
 5+>m\\r'   )ri   r   r  r   )r7   r8   r9   r:   has_attentionsr   r   r   r   r   rT   r   r   r	   r5   r>   r?   r@   s   @r%   r   r   h  s^     N   #]ll#] +,#] 
	#]  ! #]r'   r   )r   r   r   r   )0r;   r   r    r   r   activationsr   backbone_utilsr   r   modeling_outputsr	   r
   r   r   modeling_utilsr   processing_utilsr   utilsr   r   r   utils.genericr   r   utils.output_capturingr   configuration_convnextv2r   
get_loggerr7   loggerModuler   r   rB   rV   rl   r   r   r   r   r   r   r   __all__r   r'   r%   <module>r     sQ       & ! H  . & @ @ I 5 6 
		H	%BII $",, 6299 2% %0(bii (X!bii !H % % %"%O1 %OP !
/ !
 !
H )
'@ )
)
X 9](A 9]9]x ur'   