§
    9²¸j>¥  ã                   ó  — d Z ddlZddlZddlm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mZmZ ddlmZ  ej        e¦  «        Z ed¬¦  «        e G d„ de¦  «        ¦   «         ¦   «         Z ed¬¦  «        e G d„ de¦  «        ¦   «         ¦   «         Z ed¬¦  «        e G d„ de¦  «        ¦   «         ¦   «         Zd„ Zd„ Z G d„ dej         ¦  «        Z! G d„ dej         ¦  «        Z" G d„ dej         ¦  «        Z# G d„ d ej         ¦  «        Z$ G d!„ d"ej         ¦  «        Z% G d#„ d$ej         ¦  «        Z& G d%„ d&ej         ¦  «        Z' G d'„ d(ej         ¦  «        Z( G d)„ d*ej         ¦  «        Z) G d+„ d,ej         ¦  «        Z* G d-„ d.e¦  «        Z+ G d/„ d0ej         ¦  «        Z,e G d1„ d2e¦  «        ¦   «         Z-e G d3„ d4e-¦  «        ¦   «         Z. ed5¬¦  «         G d6„ d7e-¦  «        ¦   «         Z/g d8¢Z0dS )9z¢PyTorch Donut Swin Transformer model.

This implementation is identical to a regular Swin Transformer, without final layer norm on top of the final hidden
states.é    N)Ú	dataclass)Únné   )Úinitialization)ÚACT2FN)ÚGradientCheckpointingLayer)ÚPreTrainedModel)ÚModelOutputÚauto_docstringÚloggingÚ	torch_inté   )ÚDonutSwinConfigzS
    DonutSwin encoder's outputs, with potential hidden states and attentions.
    )Úcustom_introc                   ó¼   — e Zd ZU dZdZej        dz  ed<   dZe	ej        df         dz  ed<   dZ
e	ej        df         dz  ed<   dZe	ej        df         dz  ed<   dS )ÚDonutSwinEncoderOutputaí  
    reshaped_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
        Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of
        shape `(batch_size, hidden_size, height, width)`.

        Hidden-states of the model at the output of each layer plus the initial embedding outputs reshaped to
        include the spatial dimensions.
    NÚlast_hidden_state.Úhidden_statesÚ
attentionsÚreshaped_hidden_states)Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   ÚtorchÚFloatTensorÚ__annotations__r   Útupler   r   © ó    ú}/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/transformers/models/donut/modeling_donut_swin.pyr   r   %   sž   € € € € € € ðð ð 37Ð�uÔ(¨4Ñ/Ð6Ð6Ñ6Ø:>€M�5˜Ô*¨CÐ/Ô0°4Ñ7Ð>Ð>Ñ>Ø7;€J��eÔ'¨Ð,Ô-°Ñ4Ð;Ð;Ñ;ØCGÐ˜E %Ô"3°SÐ"8Ô9¸DÑ@ÐGÐGÑGÐGÐGr    r   z[
    DonutSwin model's outputs that also contains a pooling of the last hidden states.
    c                   óÚ   — 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
ej        df         dz  ed<   dZe
ej        df         dz  ed<   dZe
ej        df         dz  ed<   dS )	ÚDonutSwinModelOutputa±  
    pooler_output (`torch.FloatTensor` of shape `(batch_size, hidden_size)`, *optional*, returned when `add_pooling_layer=True` is passed):
        Average pooling of the last layer hidden-state.
    reshaped_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
        Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of
        shape `(batch_size, hidden_size, height, width)`.

        Hidden-states of the model at the output of each layer plus the initial embedding outputs reshaped to
        include the spatial dimensions.
    Nr   Úpooler_output.r   r   r   )r   r   r   r   r   r   r   r   r$   r   r   r   r   r   r    r!   r#   r#   <   s¶   € € € € € € ð	ð 	ð 37Ð�uÔ(¨4Ñ/Ð6Ð6Ñ6Ø.2€M�5Ô$ tÑ+Ð2Ð2Ñ2Ø:>€M�5˜Ô*¨CÐ/Ô0°4Ñ7Ð>Ð>Ñ>Ø7;€J��eÔ'¨Ð,Ô-°Ñ4Ð;Ð;Ñ;ØCGÐ˜E %Ô"3°SÐ"8Ô9¸DÑ@ÐGÐGÑGÐGÐGr    r#   z5
    DonutSwin outputs for image classification.
    c                   óÚ   — 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
ej        df         dz  ed<   dZe
ej        df         dz  ed<   dZe
ej        df         dz  ed<   dS )	ÚDonutSwinImageClassifierOutputa7  
    loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
        Classification (or regression if config.num_labels==1) loss.
    logits (`torch.FloatTensor` of shape `(batch_size, config.num_labels)`):
        Classification (or regression if config.num_labels==1) scores (before SoftMax).
    reshaped_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
        Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of
        shape `(batch_size, hidden_size, height, width)`.

        Hidden-states of the model at the output of each layer plus the initial embedding outputs reshaped to
        include the spatial dimensions.
    NÚlossÚlogits.r   r   r   )r   r   r   r   r'   r   r   r   r(   r   r   r   r   r   r    r!   r&   r&   V   sµ   € € € € € € ðð ð &*€Dˆ%Ô
˜dÑ
"Ð)Ð)Ñ)Ø'+€FˆEÔ Ñ$Ð+Ð+Ñ+Ø:>€M�5˜Ô*¨CÐ/Ô0°4Ñ7Ð>Ð>Ñ>Ø7;€J��eÔ'¨Ð,Ô-°Ñ4Ð;Ð;Ñ;ØCGÐ˜E %Ô"3°SÐ"8Ô9¸DÑ@ÐGÐGÑGÐGÐGr    r&   c                 óÚ   — | j         \  }}}}|                      |||z  |||z  ||¦  «        } |                      dd¦  «                             ¦   «                              d|||¦  «        }|S )z2
    Partitions the given input into windows.
    é   r   éÿÿÿÿ©ÚshapeÚviewÚ	transposeÚ
contiguous)Úinput_featureÚwindow_sizeÚ
batch_sizeÚheightÚwidthÚnum_channelsÚwindowss          r!   Úwindow_partitionr8   s   s   € ð /<Ô.AÑ+€J�˜˜|Ø!×&Ò&Ø�F˜kÑ)¨;¸ÀÑ8LÈkÐ[gñô €Mð ×%Ò% a¨Ñ+Ô+×6Ò6Ñ8Ô8×=Ò=¸bÀ+È{Ð\hÑiÔi€GØ€Nr    c                 óÜ   — | j         d         }|                      d||z  ||z  |||¦  «        } |                      dd¦  «                             ¦   «                              d|||¦  «        } | S )z?
    Merges windows to produce higher resolution features.
    r+   r*   r   r,   )r7   r2   r4   r5   r6   s        r!   Úwindow_reverser:   €   sq   € ð ”= Ô$€LØ�lŠl˜2˜v¨Ñ4°e¸{Ñ6JÈKÐYdÐfrÑsÔs€GØ×Ò  1Ñ%Ô%×0Ò0Ñ2Ô2×7Ò7¸¸FÀEÈ<ÑXÔX€GØ€Nr    c            
       ó¤   ‡ — e Zd ZdZdˆ fd„	Zdej        dededej        fd„Z	 	 dd
ej	        d	z  dej
        d	z  dedeej                 fd„Zˆ xZS )ÚDonutSwinEmbeddingszW
    Construct the patch and position embeddings. Optionally, also the mask token.
    Fc                 ó<  •— t          ¦   «                              ¦   «          t          |¦  «        | _        | j        j        }| j        j        | _        |r-t          j        t          j
        dd|j        ¦  «        ¦  «        nd | _        |j        r6t          j        t          j
        d|dz   |j        ¦  «        ¦  «        | _        nd | _        t          j        |j        ¦  «        | _        t          j        |j        ¦  «        | _        |j        | _        || _        d S )Nr   )ÚsuperÚ__init__ÚDonutSwinPatchEmbeddingsÚpatch_embeddingsÚnum_patchesÚ	grid_sizeÚ
patch_gridr   Ú	Parameterr   ÚzerosÚ	embed_dimÚ
mask_tokenÚuse_absolute_embeddingsÚposition_embeddingsÚ	LayerNormÚnormÚDropoutÚhidden_dropout_probÚdropoutÚ
patch_sizeÚconfig)ÚselfrQ   Úuse_mask_tokenrB   Ú	__class__s       €r!   r?   zDonutSwinEmbeddings.__init__�   sê   ø€ Ý‰Œ×ÒÑÔÐå 8¸Ñ @Ô @ˆÔØÔ+Ô7ˆØÔ/Ô9ˆŒØO]Ðg�"œ,¥u¤{°1°a¸Ô9IÑ'JÔ'JÑKÔKÐKÐcgˆŒàÔ)ð 	,Ý')¤|µE´KÀÀ;ÐQRÁ?ÐTZÔTdÑ4eÔ4eÑ'fÔ'fˆDÔ$Ð$à'+ˆDÔ$å”L Ô!1Ñ2Ô2ˆŒ	Ý”z &Ô"<Ñ=Ô=ˆŒØ Ô+ˆŒØˆŒˆˆr    Ú
embeddingsr4   r5   Úreturnc                 ó”  — |j         d         dz
  }| j        j         d         dz
  }t          j                             ¦   «         s||k    r||k    r| j        S | j        dd…dd…f         }| j        dd…dd…f         }|j         d         }|| j        z  }	|| j        z  }
t          |dz  ¦  «        }|                     d|||¦  «        }|                     dddd¦  «        }t          j
                             ||	|
fdd	¬
¦  «        }|                     dddd¦  «                             dd|¦  «        }t          j        ||f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.

        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   Nr+   g      à?r   r   r*   ÚbicubicF)ÚsizeÚmodeÚalign_corners©Údim)r-   rJ   r   ÚjitÚ
is_tracingrP   r   ÚreshapeÚpermuter   Ú
functionalÚinterpolater.   Úcat)rR   rU   r4   r5   rB   Únum_positionsÚclass_pos_embedÚpatch_pos_embedr]   Ú
new_heightÚ	new_widthÚsqrt_num_positionss               r!   Úinterpolate_pos_encodingz,DonutSwinEmbeddings.interpolate_pos_encoding£   sr  € ð !Ô& qÔ)¨AÑ-ˆØÔ0Ô6°qÔ9¸AÑ=ˆõ Œy×#Ò#Ñ%Ô%ð 	,¨+¸Ò*FÐ*FÈ6ÐUZÊ?È?ØÔ+Ð+àÔ2°1°1°1°b°q°b°5Ô9ˆØÔ2°1°1°1°a°b°b°5Ô9ˆàÔ˜rÔ"ˆà˜tœÑ.ˆ
Ø˜Tœ_Ñ,ˆ	å& }°cÑ'9Ñ:Ô:ÐØ)×1Ò1°!Ð5GÐI[Ð]`ÑaÔaˆØ)×1Ò1°!°Q¸¸1Ñ=Ô=ˆåœ-×3Ò3ØØ˜iÐ(ØØð	 4ñ 
ô 
ˆð *×1Ò1°!°Q¸¸1Ñ=Ô=×BÒBÀ1ÀbÈ#ÑNÔNˆåŒy˜/¨?Ð;ÀÐCÑCÔCÐCr    NÚpixel_valuesÚbool_masked_posrk   c                 óÚ  — |j         \  }}}}|                      |¦  «        \  }}	|                      |¦  «        }|                     ¦   «         \  }
}}|�R| j                             |
|d¦  «        }|                     d¦  «                             |¦  «        }|d|z
  z  ||z  z   }| j        �'|r||  	                    |||¦  «        z   }n
|| j        z   }|  
                    |¦  «        }||	fS )Nr+   g      ð?)r-   rA   rL   rY   rH   ÚexpandÚ	unsqueezeÚtype_asrJ   rk   rO   )rR   rl   rm   rk   Ú_r6   r4   r5   rU   Úoutput_dimensionsr3   Úseq_lenÚmask_tokensÚmasks                 r!   ÚforwardzDonutSwinEmbeddings.forwardË   s	  € ð *6Ô);Ñ&ˆˆ<˜ Ø(,×(=Ò(=¸lÑ(KÔ(KÑ%ˆ
Ð%Ø—Y’Y˜zÑ*Ô*ˆ
Ø!+§¢Ñ!2Ô!2Ñˆ
�G˜QàÐ&Øœ/×0Ò0°¸WÀbÑIÔIˆKà"×,Ò,¨RÑ0Ô0×8Ò8¸ÑEÔEˆDØ# s¨T¡zÑ2°[À4Ñ5GÑGˆJàÔ#Ð/Ø'ð CØ'¨$×*GÒ*GÈ
ÐTZÐ\aÑ*bÔ*bÑb�
�
à'¨$Ô*BÑB�
à—\’\ *Ñ-Ô-ˆ
àÐ,Ð,Ð,r    )F©NF)r   r   r   r   r?   r   ÚTensorÚintrk   r   Ú
BoolTensorÚboolr   rw   Ú__classcell__©rT   s   @r!   r<   r<   ‹   så   ø€ € € € € ðð ðð ð ð ð ð ð&&D°5´<ð &DÈð &DÐUXð &DÐ]bÔ]ið &Dð &Dð &Dð &DðV 48Ø).ð	-ð -àÔ'¨$Ñ.ð-ð Ô)¨DÑ0ð-ð #'ð	-ð
 
ˆuŒ|Ô	ð-ð -ð -ð -ð -ð -ð -ð -r    r<   c                   ón   ‡ — e Zd ZdZˆ fd„Zd„ Zdej        dz  deej	        ee
         f         fd„Zˆ xZS )r@   zì
    This class turns `pixel_values` of shape `(batch_size, num_channels, height, width)` into the initial
    `hidden_states` (patch embeddings) of shape `(batch_size, seq_length, hidden_size)` to be consumed by a
    Transformer.
    c                 óþ  •— t          ¦   «                              ¦   «          |j        |j        }}|j        |j        }}t          |t          j        j	        ¦  «        r|n||f}t          |t          j        j	        ¦  «        r|n||f}|d         |d         z  |d         |d         z  z  }|| _        || _
        |d         |d         z  |d         |d         z  f| _        t          j        ||||¬¦  «        | _        d S )Nr   r   )Úkernel_sizeÚstride)r>   r?   Ú
image_sizerP   r6   rG   Ú
isinstanceÚcollectionsÚabcÚIterablerB   rC   r   ÚConv2dÚ
projection)rR   rQ   rƒ   rP   r6   Úhidden_sizerB   rT   s          €r!   r?   z!DonutSwinPatchEmbeddings.__init__ï   sù   ø€ Ý‰Œ×ÒÑÔÐØ!'Ô!2°FÔ4E�Jˆ
Ø$*Ô$7¸Ô9I�kˆÝ#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ý#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ø! !”}¨
°1¬Ñ5¸*ÀQ¼-È:ÐVWÌ=Ñ:XÑYˆØ$ˆŒØ&ˆÔØ$ Qœ-¨:°a¬=Ñ8¸*ÀQ¼-È:ÐVWÌ=Ñ:XÐYˆŒåœ) L°+È:Ð^hÐiÑiÔiˆŒˆˆr    c                 óZ  — || j         d         z  dk    r@d| j         d         || j         d         z  z
  f}t          j                             ||¦  «        }|| j         d         z  dk    rBddd| j         d         || j         d         z  z
  f}t          j                             ||¦  «        }|S )z9Pad pixel_values to be divisible by patch_size if needed.r   r   )rP   r   rb   Úpad)rR   rl   r4   r5   Ú
pad_valuess        r!   Ú	maybe_padz"DonutSwinPatchEmbeddings.maybe_padü   s­   € à�4”? 1Ô%Ñ%¨Ò*Ð*Ø˜Tœ_¨QÔ/°%¸$¼/È!Ô:LÑ2LÑLÐMˆJÝœ=×,Ò,¨\¸:ÑFÔFˆLØ�D”O AÔ&Ñ&¨!Ò+Ð+Ø˜Q  4¤?°1Ô#5¸ÀÄÐQRÔASÑ8SÑ#SÐTˆJÝœ=×,Ò,¨\¸:ÑFÔFˆLØÐr    rl   NrV   c                 óì   — |j         \  }}}}|                      |||¦  «        }|                      |¦  «        }|j         \  }}}}||f}|                     d¦  «                             dd¦  «        }||fS )Nr*   r   )r-   rŽ   r‰   Úflattenr/   )rR   rl   rr   r6   r4   r5   rU   rs   s           r!   rw   z DonutSwinPatchEmbeddings.forward  sƒ   € Ø)5Ô);Ñ&ˆˆ<˜ à—~’~ l°F¸EÑBÔBˆØ—_’_ \Ñ2Ô2ˆ
Ø(Ô.Ñˆˆ1ˆf�eØ# U˜OÐØ×'Ò'¨Ñ*Ô*×4Ò4°Q¸Ñ:Ô:ˆ
àÐ,Ð,Ð,r    )r   r   r   r   r?   rŽ   r   r   r   ry   rz   rw   r}   r~   s   @r!   r@   r@   è   s“   ø€ € € € € ðð ðjð jð jð jð jðð ð ð	- EÔ$5¸Ñ$<ð 	-ÀÀuÄ|ÐUZÐ[^ÔU_ÐG_ÔA`ð 	-ð 	-ð 	-ð 	-ð 	-ð 	-ð 	-ð 	-r    r@   c                   ó�   ‡ — e Zd ZdZdeddfˆ fd„Zdej        dededej        fd	„Zdej        d
e	eef         dej        fd„Z
ˆ xZS )ÚDonutSwinPatchMergingzd
    Patch Merging Layer.

    Args:
        dim (`int`):
            Number of input channels.
    r]   rV   Nc                 óÄ   •— t          ¦   «                              ¦   «          t          j        d|z  d|z  d¬¦  «        | _        t          j        d|z  ¦  «        | _        d S )Né   r*   F©Úbias)r>   r?   r   ÚLinearÚ	reductionrK   rL   )rR   r]   rT   s     €r!   r?   zDonutSwinPatchMerging.__init__  sR   ø€ Ý‰Œ×ÒÑÔÐÝœ 1 s¡7¨A°©G¸%Ð@Ñ@Ô@ˆŒÝ”L  S¡Ñ)Ô)ˆŒ	ˆ	ˆ	r    r1   r4   r5   c           
      ó‚   — |dz  dk    s	|dz  dk    r,t           j                             |ddd|dz  d|dz  f¦  «        }|S )zPPad input feature map to be divisible by 2 in both spatial dimensions if needed.r*   r   r   )r   rb   rŒ   )rR   r1   r4   r5   s       r!   rŽ   zDonutSwinPatchMerging.maybe_pad!  sQ   € à�Q‰J˜!ŠOˆO ¨¡¨a¢ ÝœM×-Ò-¨m¸aÀÀAÀuÈqÁyÐRSÐU[Ð^_ÑU_Ð=`ÑaÔaˆMØÐr    Úinput_dimensionsc                 ól  ‡— |\  }}‰j         \  }}}‰                     ||||¦  «        Š|                      ‰||¦  «        Št          j        ˆfd„t          d¦  «        D ¦   «         d¬¦  «        Š‰                     |dd|z  ¦  «        Š|                      ‰¦  «        Š|                      ‰¦  «        Š‰S )Nc           	      ó`   •— g | ]*}t          d ¦  «        D ]}‰dd…|dd …|dd …dd…f         ‘ŒŒ+S )r*   N)Úrange)Ú.0ÚcolÚrowr1   s      €r!   ú
<listcomp>z1DonutSwinPatchMerging.forward.<locals>.<listcomp>1  sR   ø€ ÐYÐYÐY°SÕPUÐVWÑPXÔPXÐYÐYÈˆ]˜1˜1˜1˜c˜f 1˜f c f¨1 f¨a¨a¨aÐ/Ô0ÐYÐYÐYÐYr    r*   r+   r\   r”   )r-   r.   rŽ   r   rd   r�   rL   r˜   )rR   r1   rš   r4   r5   r3   r]   r6   s    `      r!   rw   zDonutSwinPatchMerging.forward'  sÅ   ø€ Ø(‰ˆ�à(5Ô(;Ñ%ˆ
�C˜à%×*Ò*¨:°v¸uÀlÑSÔSˆàŸš }°f¸eÑDÔDˆåœ	ØYÐYÐYÐY½EÀ!¹H¼HÐYÑYÔYÐ_að
ñ 
ô 
ˆð &×*Ò*¨:°r¸1¸|Ñ;KÑLÔLˆàŸ	š	 -Ñ0Ô0ˆØŸš }Ñ5Ô5ˆàÐr    )r   r   r   r   rz   r?   r   ry   rŽ   r   rw   r}   r~   s   @r!   r’   r’     sÃ   ø€ € € € € ðð ð*˜Cð * Dð *ð *ð *ð *ð *ð *ð
 u¤|ð ¸Sð Èð ÐQVÔQ]ð ð ð ð ð U¤\ð ÀUÈ3ÐPSÈ8Ä_ð ÐY^ÔYeð ð ð ð ð ð ð ð r    r’   c            
       óx   ‡ — e Zd Zˆ fd„Z	 	 d
dej        dej        dz  dedz  deej                 fd„Z	d	„ Z
ˆ xZS )ÚDonutSwinSelfAttentionc                 ót  •— t          ¦   «                              ¦   «          ||z  dk    rt          d|› d|› d�¦  «        ‚|| _        t	          ||z  ¦  «        | _        | j        | j        z  | _        t          |t          j	        j
        ¦  «        r|n||f| _        t          j        t          j        d| j        d         z  dz
  d| j        d         z  dz
  z  |¦  «        ¦  «        | _        t          j        |                      ¦   «         ¦  «        | _        t          j        | j        | j        |j        ¬¦  «        | _        t          j        | j        | j        |j        ¬¦  «        | _        t          j        | j        | j        |j        ¬¦  «        | _        t          j        |j        ¦  «        | _        d S )Nr   zThe hidden size (z6) is not a multiple of the number of attention heads (ú)r*   r   r•   )r>   r?   Ú
ValueErrorÚnum_attention_headsrz   Úattention_head_sizeÚall_head_sizer„   r…   r†   r‡   r2   r   rE   r   rF   Úrelative_position_bias_tableÚBufferÚcreate_relative_position_indexÚrelative_position_indexr—   Úqkv_biasÚqueryÚkeyÚvaluerM   Úattention_probs_dropout_probrO   ©rR   rQ   r]   Ú	num_headsr2   rT   s        €r!   r?   zDonutSwinSelfAttention.__init__=  s”  ø€ Ý‰Œ×ÒÑÔÐØ�‰?˜aÒÐÝØk CÐkÐkÐ_hÐkÐkÐkñô ð ð $-ˆÔ Ý#& s¨Y¡Ñ#7Ô#7ˆÔ Ø!Ô5¸Ô8PÑPˆÔå% kµ;´?Ô3KÑLÔLÐlˆKˆKÐS^Ð`kÐRlð 	Ôõ -/¬LÝŒK˜˜TÔ-¨aÔ0Ñ0°1Ñ4¸¸TÔ=MÈaÔ=PÑ9PÐSTÑ9TÑUÐW`ÑaÔañ-
ô -
ˆÔ)õ (*¤y°×1TÒ1TÑ1VÔ1VÑ'WÔ'WˆÔ$å”Y˜tÔ1°4Ô3EÈFÌOÐ\Ñ\Ô\ˆŒ
Ý”9˜TÔ/°Ô1CÈ&Ì/ÐZÑZÔZˆŒÝ”Y˜tÔ1°4Ô3EÈFÌOÐ\Ñ\Ô\ˆŒ
å”z &Ô"EÑFÔFˆŒˆˆr    NFr   Úattention_maskÚoutput_attentionsrV   c                 óâ  — |j         \  }}}||d| j        f}|                      |¦  «                             |¦  «                             dd¦  «        }|                      |¦  «                             |¦  «                             dd¦  «        }	|                      |¦  «                             |¦  «                             dd¦  «        }
t          j        ||	                     dd¦  «        ¦  «        }|t          j
        | j        ¦  «        z  }| j        | j                             d¦  «                 }|                     | j        d         | j        d         z  | j        d         | j        d         z  d¦  «        }|                     ddd¦  «                             ¦   «         }||                     d¦  «        z   }|�v|j         d         }|                     ||z  || j        ||¦  «        }||                     d¦  «                             d¦  «        z   }|                     d| j        ||¦  «        }t$          j                             |d¬¦  «        }|                      |¦  «        }t          j        ||
¦  «        }|                     dddd¦  «                             ¦   «         }|                     ¦   «         d d…         | j        fz   }|                     |¦  «        }|r||fn|f}|S )Nr+   r   r*   éþÿÿÿr   r\   r   )r-   r¨   r¯   r.   r/   r°   r±   r   ÚmatmulÚmathÚsqrtrª   r­   r2   ra   r0   rp   r§   r   rb   ÚsoftmaxrO   rY   r©   )rR   r   rµ   r¶   r3   r]   r6   Úhidden_shapeÚquery_layerÚ	key_layerÚvalue_layerÚattention_scoresÚrelative_position_biasÚ
mask_shapeÚattention_probsÚcontext_layerÚnew_context_layer_shapeÚoutputss                     r!   rw   zDonutSwinSelfAttention.forwardW  sÛ  € ð )6Ô(;Ñ%ˆ
�C˜Ø" C¨¨TÔ-EÐFˆà—j’j Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆØ—H’H˜]Ñ+Ô+×0Ò0°Ñ>Ô>×HÒHÈÈAÑNÔNˆ	Ø—j’j Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆõ !œ<¨°Y×5HÒ5HÈÈRÑ5PÔ5PÑQÔQÐà+­d¬i¸Ô8PÑ.QÔ.QÑQÐà!%Ô!BÀ4ÔC_×CdÒCdÐegÑChÔChÔ!iÐØ!7×!<Ò!<ØÔ˜QÔ $Ô"2°1Ô"5Ñ5°tÔ7GÈÔ7JÈTÔM]Ð^_ÔM`Ñ7`Ðbdñ"
ô "
Ðð "8×!?Ò!?ÀÀ1ÀaÑ!HÔ!H×!SÒ!SÑ!UÔ!UÐØ+Ð.D×.NÒ.NÈqÑ.QÔ.QÑQÐàÐ%à'Ô-¨aÔ0ˆJØ/×4Ò4Ø˜jÑ(¨*°dÔ6NÐPSÐUXñ ô  Ðð  0°.×2JÒ2JÈ1Ñ2MÔ2M×2WÒ2WÐXYÑ2ZÔ2ZÑZÐØ/×4Ò4°R¸Ô9QÐSVÐX[Ñ\Ô\Ðõ œ-×/Ò/Ð0@ÀbÐ/ÑIÔIˆð Ÿ,š, Ñ7Ô7ˆåœ _°kÑBÔBˆØ%×-Ò-¨a°°A°qÑ9Ô9×DÒDÑFÔFˆØ"/×"4Ò"4Ñ"6Ô"6°s¸°sÔ";¸tÔ?QÐ>SÑ"SÐØ%×*Ò*Ð+BÑCÔCˆà6GÐ]�= /Ð2Ð2ÈmÐM]ˆàˆr    c                 óž  — t          j        | j        d         ¦  «        }t          j        | j        d         ¦  «        }t          j        t          j        ||gd¬¦  «        ¦  «        }t          j        |d¦  «        }|d d …d d …d f         |d d …d d d …f         z
  }|                     ddd¦  «                             ¦   «         }|d d …d d …dfxx         | j        d         dz
  z  cc<   |d d …d d …dfxx         | j        d         dz
  z  cc<   |d d …d d …dfxx         d| j        d         z  dz
  z  cc<   |                     d¦  «        }|S )Nr   r   Úij)Úindexingr*   r+   )	r   Úaranger2   ÚstackÚmeshgridr�   ra   r0   Úsum)rR   Úcoords_hÚcoords_wÚcoordsÚcoords_flattenÚrelative_coordsr­   s          r!   r¬   z5DonutSwinSelfAttention.create_relative_position_indexŠ  s{  € å”< Ô 0°Ô 3Ñ4Ô4ˆÝ”< Ô 0°Ô 3Ñ4Ô4ˆÝ”�Uœ^¨X°xÐ,@È4ÐPÑPÔPÑQÔQˆÝœ v¨qÑ1Ô1ˆØ(¨¨¨¨A¨A¨A¨t¨Ô4°~ÀaÀaÀaÈÈqÈqÈqÀjÔ7QÑQˆØ)×1Ò1°!°Q¸Ñ:Ô:×EÒEÑGÔGˆØ˜˜˜˜1˜1˜1˜a˜Ð Ð Ô  DÔ$4°QÔ$7¸!Ñ$;Ñ;Ð Ð Ñ Ø˜˜˜˜1˜1˜1˜a˜Ð Ð Ô  DÔ$4°QÔ$7¸!Ñ$;Ñ;Ð Ð Ñ Ø˜˜˜˜1˜1˜1˜a˜Ð Ð Ô  A¨Ô(8¸Ô(;Ñ$;¸aÑ$?Ñ?Ð Ð Ñ Ø"1×"5Ò"5°bÑ"9Ô"9ÐØ&Ð&r    rx   )r   r   r   r?   r   ry   r   r|   r   rw   r¬   r}   r~   s   @r!   r£   r£   <  s©   ø€ € € € € ðGð Gð Gð Gð Gð: 48Ø).ð	1ð 1à”|ð1ð Ô)¨DÑ0ð1ð   $™;ð	1ð
 
ˆuŒ|Ô	ð1ð 1ð 1ð 1ðf'ð 'ð 'ð 'ð 'ð 'ð 'r    r£   c                   óP   ‡ — e Zd Zˆ fd„Zdej        dej        dej        fd„Zˆ xZS )ÚDonutSwinSelfOutputc                 ó¸   •— t          ¦   «                              ¦   «          t          j        ||¦  «        | _        t          j        |j        ¦  «        | _        d S ©N)r>   r?   r   r—   ÚdenserM   r²   rO   ©rR   rQ   r]   rT   s      €r!   r?   zDonutSwinSelfOutput.__init__›  sD   ø€ Ý‰Œ×ÒÑÔÐÝ”Y˜s CÑ(Ô(ˆŒ
Ý”z &Ô"EÑFÔFˆŒˆˆr    r   Úinput_tensorrV   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r×   ©rØ   rO   )rR   r   rÚ   s      r!   rw   zDonutSwinSelfOutput.forward   s*   € ØŸ
š
 =Ñ1Ô1ˆØŸš ]Ñ3Ô3ˆàÐr    ©r   r   r   r?   r   ry   rw   r}   r~   s   @r!   rÕ   rÕ   š  sn   ø€ € € € € ðGð Gð Gð Gð Gð
 U¤\ð ÀÄð ÐRWÔR^ð ð ð ð ð ð ð ð r    rÕ   c            
       ór   ‡ — e Zd Zˆ fd„Z	 	 d	dej        dej        dz  dedz  deej                 fd„Z	ˆ xZ
S )
ÚDonutSwinAttentionc                 ó    •— t          ¦   «                              ¦   «          t          ||||¦  «        | _        t	          ||¦  «        | _        d S r×   )r>   r?   r£   rR   rÕ   Úoutputr³   s        €r!   r?   zDonutSwinAttention.__init__©  sC   ø€ Ý‰Œ×ÒÑÔÐÝ*¨6°3¸	À;ÑOÔOˆŒ	Ý)¨&°#Ñ6Ô6ˆŒˆˆr    NFr   rµ   r¶   rV   c                 óˆ   — |                       |||¦  «        }|                      |d         |¦  «        }|f|dd …         z   }|S )Nr   r   )rR   rá   )rR   r   rµ   r¶   Úself_outputsÚattention_outputrÇ   s          r!   rw   zDonutSwinAttention.forward®  sM   € ð —y’y °Ð@QÑRÔRˆØŸ;š; |°A¤¸ÑFÔFÐØ#Ð%¨°Q°R°RÔ(8Ñ8ˆØˆr    rx   )r   r   r   r?   r   ry   r   r|   r   rw   r}   r~   s   @r!   rß   rß   ¨  s”   ø€ € € € € ð7ð 7ð 7ð 7ð 7ð 48Ø).ð		ð 	à”|ð	ð Ô)¨DÑ0ð	ð   $™;ð		ð
 
ˆuŒ|Ô	ð	ð 	ð 	ð 	ð 	ð 	ð 	ð 	r    rß   c                   óB   ‡ — e Zd Zˆ fd„Zdej        dej        fd„Zˆ xZS )ÚDonutSwinIntermediatec                 ó$  •— t          ¦   «                              ¦   «          t          j        |t	          |j        |z  ¦  «        ¦  «        | _        t          |j        t          ¦  «        rt          |j                 | _        d S |j        | _        d S r×   )r>   r?   r   r—   rz   Ú	mlp_ratiorØ   r„   Ú
hidden_actÚstrr   Úintermediate_act_fnrÙ   s      €r!   r?   zDonutSwinIntermediate.__init__¼  sx   ø€ Ý‰Œ×ÒÑÔÐÝ”Y˜s¥C¨Ô(8¸3Ñ(>Ñ$?Ô$?Ñ@Ô@ˆŒ
Ý�fÔ'­Ñ-Ô-ð 	9Ý'-¨fÔ.?Ô'@ˆDÔ$Ð$Ð$à'-Ô'8ˆDÔ$Ð$Ð$r    r   rV   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r×   )rØ   rë   ©rR   r   s     r!   rw   zDonutSwinIntermediate.forwardÄ  s,   € ØŸ
š
 =Ñ1Ô1ˆØ×0Ò0°Ñ?Ô?ˆØÐr    rÝ   r~   s   @r!   ræ   ræ   »  s^   ø€ € € € € ð9ð 9ð 9ð 9ð 9ð U¤\ð °e´lð ð ð ð ð ð ð ð r    ræ   c                   óB   ‡ — e Zd Zˆ fd„Zdej        dej        fd„Zˆ xZS )ÚDonutSwinOutputc                 óâ   •— t          ¦   «                              ¦   «          t          j        t	          |j        |z  ¦  «        |¦  «        | _        t          j        |j        ¦  «        | _	        d S r×   )
r>   r?   r   r—   rz   rè   rØ   rM   rN   rO   rÙ   s      €r!   r?   zDonutSwinOutput.__init__Ì  sT   ø€ Ý‰Œ×ÒÑÔÐÝ”Y�s 6Ô#3°cÑ#9Ñ:Ô:¸CÑ@Ô@ˆŒ
Ý”z &Ô"<Ñ=Ô=ˆŒˆˆr    r   rV   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r×   rÜ   rí   s     r!   rw   zDonutSwinOutput.forwardÑ  s*   € ØŸ
š
 =Ñ1Ô1ˆØŸš ]Ñ3Ô3ˆØÐr    rÝ   r~   s   @r!   rï   rï   Ë  s^   ø€ € € € € ð>ð >ð >ð >ð >ð
 U¤\ð °e´lð ð ð ð ð ð ð ð r    rï   c                   ó^   ‡ — e Zd ZdZd
deddfˆ fd„Zdej        dej        fd„Zde	fd	„Z
ˆ xZS )ÚDonutDropPathzÏStochastic 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>`_.
    ç        Ú	drop_probrV   Nc                 óV   •— t          ¦   «                              ¦   «          || _        d S r×   )r>   r?   rõ   )rR   rõ   rT   s     €r!   r?   zDonutDropPath.__init__ß  s$   ø€ Ý‰Œ×ÒÑÔÐØ"ˆŒˆˆr    r   c                 ó  — | j         dk    s| j        s|S d| j         z
  }|j        d         fd|j        dz
  z  z   }t	          j        ||j        |j        ¬¦  «        }t	          j        ||z   ¦  «        }| 	                    |¦  «        |z  S )Nrô   r   r   )r   ©ÚdtypeÚdevice)
rõ   Útrainingr-   Úndimr   Úrandrù   rú   ÚfloorÚdiv)rR   r   Ú	keep_probr-   Úrandom_tensors        r!   rw   zDonutDropPath.forwardã  s“   € ØŒ>˜SÒ Ð ¨¬Ð Ø Ð Ø˜œÑ&ˆ	ØÔ$ QÔ'Ð)¨D°MÔ4FÈÑ4JÑ,KÑKˆÝœ
 5°Ô0CÈMÔL`ÐaÑaÔaˆÝœ M°IÑ$=Ñ>Ô>ˆØ× Ò  Ñ+Ô+¨mÑ;Ð;r    c                 ó   — d| j         › �S )Nzp=)rõ   ©rR   s    r!   Ú
extra_reprzDonutDropPath.extra_reprì  s   € Ø$�D”NÐ$Ð$Ð$r    )rô   )r   r   r   r   Úfloatr?   r   ry   rw   rê   r  r}   r~   s   @r!   ró   ró   Ø  s›   ø€ € € € € ðð ð#ð # %ð #°$ð #ð #ð #ð #ð #ð #ð< U¤\ð <°e´lð <ð <ð <ð <ð%˜Cð %ð %ð %ð %ð %ð %ð %ð %r    ró   c                   óØ   ‡ — e Zd Zdˆ fd„	Zd„ Zdededej        dej        d	ej	        d
z  f
d„Z
d„ Z	 	 ddej	        deeef         ded
z  ded
z  d	eej	        ej	        f         f
d„Zˆ xZS )ÚDonutSwinLayerrô   r   c                 óü  •— t          ¦   «                              ¦   «          |j        | _        || _        |j        | _        || _        t          j        ||j        ¬¦  «        | _	        t          |||| j        ¬¦  «        | _        |dk    rt          |¦  «        nt          j        ¦   «         | _        t          j        ||j        ¬¦  «        | _        t!          ||¦  «        | _        t%          ||¦  «        | _        d S )N)Úeps)r2   rô   )r>   r?   Úchunk_size_feed_forwardÚ
shift_sizer2   Úinput_resolutionr   rK   Úlayer_norm_epsÚlayernorm_beforerß   Ú	attentionró   ÚIdentityÚ	drop_pathÚlayernorm_afterræ   Úintermediaterï   rá   )rR   rQ   r]   r  r´   Údrop_path_rater  rT   s          €r!   r?   zDonutSwinLayer.__init__ò  sÝ   ø€ Ý‰Œ×ÒÑÔÐØ'-Ô'EˆÔ$Ø$ˆŒØ!Ô-ˆÔØ 0ˆÔÝ "¤¨S°fÔ6KÐ LÑ LÔ LˆÔÝ+¨F°C¸ÐPTÔP`ÐaÑaÔaˆŒØ:HÈ3Ò:NÐ:N� ~Ñ6Ô6Ð6ÕTVÔT_ÑTaÔTaˆŒÝ!œ|¨C°VÔ5JÐKÑKÔKˆÔÝ1°&¸#Ñ>Ô>ˆÔÝ% f¨cÑ2Ô2ˆŒˆˆr    c                 ó  — t          |¦  «        | j        k    rnt          d¦  «        | _        t          j                             ¦   «         r&t	          j         t	          j        |¦  «        ¦  «        nt          |¦  «        | _        d S d S ©Nr   )Úminr2   r   r  r   r^   r_   Útensor)rR   r  s     r!   Úset_shift_and_window_sizez(DonutSwinLayer.set_shift_and_window_sizeÿ  sv   € ÝÐÑ Ô  DÔ$4Ò4Ð4å'¨™lœlˆDŒOå=B¼Y×=QÒ=QÑ=SÔ=SÐn•”	�%œ,Ð'7Ñ8Ô8Ñ9Ô9Ð9ÕY\Ð]mÑYnÔYnð ÔÐÐð 5Ð4r    r4   r5   rù   rú   rV   Nc                 ó  — | j         dk    rdS t          j        ||¬¦  «        }t          j        ||¬¦  «        }||| j        z
  k                         ¦   «         ||| j         z
  k                         ¦   «         z   }||| j        z
  k                         ¦   «         ||| j         z
  k                         ¦   «         z   }|ddd…ddf         dz  |dddd…df         z                        |¦  «        }	t          |	| j        ¦  «        }
|
                     d| j        | j        z  ¦  «        }
|
                     d¦  «        |
                     d¦  «        z
  }| 	                    |dk    d¦  «         	                    |dk    d	¦  «        }|S )
uµ  Build the cyclic-shift attention mask for shifted-window MSA; returns None when shift_size is 0.

        Each (h, w) position belongs to one of 9 cyclic-shift regions (3 along each axis), encoded
        as ``h_region * 3 + w_region``. Regions per axis:
        - 0: indices ``[0, axis - window_size)``
        - 1: indices ``[axis - window_size, axis - shift_size)``
        - 2: indices ``[axis - shift_size, axis)``
        Implementation note: a single arithmetic pass on `torch.arange` (two comparisons +
        broadcast add) replaces the original 9-iteration nested-Python-loop slice-assignment â€”
        fully vectorised, no per-cell host-side scatter, no GPUâ†”host sync.
        r   N©rú   r   r+   r   r*   g      YÀrô   )
r  r   rË   r2   ÚlongÚtor8   r.   rp   Úmasked_fill)rR   r4   r5   rù   rú   Úh_idxÚw_idxÚh_regionÚw_regionÚimg_maskÚmask_windowsÚ	attn_masks               r!   Úget_attn_maskzDonutSwinLayer.get_attn_mask  sˆ  € ð Œ?˜aÒÐØ�4Ý”˜V¨FÐ3Ñ3Ô3ˆÝ”˜U¨6Ð2Ñ2Ô2ˆØ˜V dÔ&6Ñ6Ò6×<Ò<Ñ>Ô>À%È6ÐTXÔTcÑKcÒBc×AiÒAiÑAkÔAkÑkˆØ˜U TÔ%5Ñ5Ò5×;Ò;Ñ=Ô=ÀÈ%ÐRVÔRaÑJaÒAa×@gÒ@gÑ@iÔ@iÑiˆØ˜T 1 1 1 d¨DÐ0Ô1°AÑ5¸ÀÀtÈQÈQÈQÐPTÐATÔ8UÑU×YÒYÐZ_Ñ`Ô`ˆÝ'¨°$Ô2BÑCÔCˆØ#×(Ò(¨¨TÔ-=ÀÔ@PÑ-PÑQÔQˆØ ×*Ò*¨1Ñ-Ô-°×0FÒ0FÀqÑ0IÔ0IÑIˆ	Ø×)Ò)¨)°qª.¸&ÑAÔA×MÒMÈiÐ[\ÊnÐ^aÑbÔbˆ	ØÐr    c                 óÂ   — | j         || j         z  z
  | j         z  }| j         || j         z  z
  | j         z  }ddd|d|f}t          j                             ||¦  «        }||fS r  )r2   r   rb   rŒ   )rR   r   r4   r5   Ú	pad_rightÚ
pad_bottomr�   s          r!   rŽ   zDonutSwinLayer.maybe_pad!  sp   € ØÔ%¨°Ô0@Ñ(@Ñ@ÀDÔDTÑTˆ	ØÔ&¨°$Ô2BÑ)BÑBÀdÔFVÑVˆ
Ø˜˜A˜y¨!¨ZÐ8ˆ
Ýœ×)Ò)¨-¸ÑDÔDˆØ˜jÐ(Ð(r    Fr   rš   r¶   Úalways_partitionc                 óÔ  — |s|                       |¦  «         n	 |\  }}|                     ¦   «         \  }}}	|}
|                      |¦  «        }|                     ||||	¦  «        }|                      |||¦  «        \  }}|j        \  }}}}| j        dk    r&t          j        || j         | j         fd¬¦  «        }n|}t          || j
        ¦  «        }|                     d| j
        | j
        z  |	¦  «        }|                      |||j        |j        ¬¦  «        }|                      |||¬¦  «        }|d         }|                     d| j
        | j
        |	¦  «        }t          || j
        ||¦  «        }| j        dk    r$t          j        || j        | j        fd¬¦  «        }n|}|d         dk    p|d         dk    }|r&|d d …d |…d |…d d …f                              ¦   «         }|                     |||z  |	¦  «        }|
|                      |¦  «        z   }|                      |¦  «        }|                      |¦  «        }||                      |¦  «        z   }|r
||d	         fn|f}|S )
Nr   )r   r*   )ÚshiftsÚdimsr+   rø   )r¶   r   é   r   )r  rY   r  r.   rŽ   r-   r  r   Úrollr8   r2   r&  rù   rú   r  r:   r0   r  r  r  rá   )rR   r   rš   r¶   r*  r4   r5   r3   rr   ÚchannelsÚshortcutr�   Ú
height_padÚ	width_padÚshifted_hidden_statesÚhidden_states_windowsr%  Úattention_outputsrä   Úattention_windowsÚshifted_windowsÚ
was_paddedÚlayer_outputÚlayer_outputss                           r!   rw   zDonutSwinLayer.forward(  sÍ  € ð  ð 	Ø×*Ò*Ð+;Ñ<Ô<Ð<Ð<àØ(‰ˆ�Ø"/×"4Ò"4Ñ"6Ô"6Ñˆ
�A�xØ ˆà×-Ò-¨mÑ<Ô<ˆà%×*Ò*¨:°v¸uÀhÑOÔOˆð %)§N¢N°=À&È%Ñ$PÔ$PÑ!ˆ�zà&3Ô&9Ñ#ˆˆ:�y !àŒ?˜QÒÐÝ$)¤J¨}ÀtÄÐFVÐY]ÔYhÐXhÐEiÐpvÐ$wÑ$wÔ$wÐ!Ð!à$1Ð!õ !1Ð1FÈÔHXÑ YÔ YÐØ 5× :Ò :¸2¸tÔ?OÐRVÔRbÑ?bÐdlÑ mÔ mÐØ×&Ò&Ø˜	¨Ô)<ÐEZÔEað 'ñ 
ô 
ˆ	ð !ŸNšNÐ+@À)Ð_p˜NÑqÔqÐà,¨QÔ/Ðà,×1Ò1°"°dÔ6FÈÔHXÐZbÑcÔcÐÝ(Ð):¸DÔ<LÈjÐZcÑdÔdˆð Œ?˜QÒÐÝ %¤
¨?ÀDÄOÐUYÔUdÐCeÐlrÐ sÑ sÔ sÐÐà /Ðà ”] QÒ&Ð;¨*°Q¬-¸!Ò*;ˆ
Øð 	VØ 1°!°!°!°W°f°W¸f¸u¸fÀaÀaÀaÐ2GÔ H× SÒ SÑ UÔ UÐà-×2Ò2°:¸vÈ¹~ÈxÑXÔXÐà  4§>¢>Ð2CÑ#DÔ#DÑDˆà×+Ò+¨MÑ:Ô:ˆØ×(Ò(¨Ñ6Ô6ˆØ$ t§{¢{°<Ñ'@Ô'@Ñ@ˆà@QÐf˜Ð'8¸Ô';Ð<Ð<ÐXdÐWfˆØÐr    )rô   r   ©FF)r   r   r   r?   r  rz   r   rù   rú   ry   r&  rŽ   r   r|   rw   r}   r~   s   @r!   r  r  ñ  s  ø€ € € € € ð3ð 3ð 3ð 3ð 3ð 3ðð ð ð Cð °ð ¸E¼Kð ÐQVÔQ]ð ÐbgÔbnÐquÑbuð ð ð ð ð2)ð )ð )ð */Ø(-ð>ð >à”|ð>ð    S œ/ð>ð   $™;ð	>ð
  ™+ð>ð 
ˆuŒ|˜Uœ\Ð)Ô	*ð>ð >ð >ð >ð >ð >ð >ð >r    r  c                   ó|   ‡ — e Zd Zˆ fd„Z	 	 d
dej        deeef         dedz  dedz  deej                 f
d	„Z	ˆ xZ
S )ÚDonutSwinStagec                 ó  •‡‡‡‡‡— t          ¦   «                              ¦   «          ‰| _        ‰| _        t	          j        ˆˆˆˆˆfd„t          |¦  «        D ¦   «         ¦  «        | _        |� |‰¬¦  «        | _        nd | _        d| _	        d S )Nc                 ól   •— g | ]0}t          ‰‰‰‰‰|         |d z  dk    rdn	‰j        d z  ¬¦  «        ‘Œ1S )r*   r   )rQ   r]   r  r´   r  r  )r  r2   )rž   ÚirQ   r]   r  r  r´   s     €€€€€r!   r¡   z+DonutSwinStage.__init__.<locals>.<listcomp>p  sh   ø€ ð 
ð 
ð 
ð õ Ø!ØØ%5Ø'Ø#,¨Q¤<Ø%&¨¡U¨a¢Z Z˜q˜q°fÔ6HÈAÑ6Mðñ ô ð
ð 
ð 
r    r\   F)
r>   r?   rQ   r]   r   Ú
ModuleListr�   ÚblocksÚ
downsampleÚpointing)	rR   rQ   r]   r  Údepthr´   r  rD  rT   s	    ``` `` €r!   r?   zDonutSwinStage.__init__k  s±   øøøøøø€ Ý‰Œ×ÒÑÔÐØˆŒØˆŒÝ”mð
ð 
ð 
ð 
ð 
ð 
ð 
ð 
õ ˜u™œð
ñ 
ô 
ñ
ô 
ˆŒð Ð!Ø(˜j¨SÐ1Ñ1Ô1ˆDŒOˆOà"ˆDŒOàˆŒˆˆr    Fr   rš   r¶   Nr*  rV   c                 ó  — |\  }}t          | j        ¦  «        D ]\  }} |||||¦  «        }	|	d         }Œ|}
| j        �-|dz   dz  |dz   dz  }}||||f}|                      |
|¦  «        }n||||f}||
|f}|r||	dd …         z  }|S )Nr   r   r*   )Ú	enumeraterC  rD  )rR   r   rš   r¶   r*  r4   r5   rA  Úlayer_moduler;  Ú!hidden_states_before_downsamplingÚheight_downsampledÚwidth_downsampledrs   Ústage_outputss                  r!   rw   zDonutSwinStage.forward…  sá   € ð )‰ˆ�Ý(¨¬Ñ5Ô5ð 	-ð 	-‰OˆAˆ|Ø(˜L¨Ð8HÐJ[Ð]mÑnÔnˆMà)¨!Ô,ˆMˆMà,9Ð)ØŒ?Ð&Ø5;¸a±ZÀAÑ4EÈÐPQÉ	ÐVWÑGWÐ 1ÐØ!'¨Ð0BÐDUÐ VÐØ ŸOšOÐ,MÐO_Ñ`Ô`ˆMˆMà!'¨°¸Ð >Ðà&Ð(IÐK\Ð]ˆàð 	/Ø˜]¨1¨2¨2Ô.Ñ.ˆMØÐr    r<  )r   r   r   r?   r   ry   r   rz   r|   rw   r}   r~   s   @r!   r>  r>  j  s¢   ø€ € € € € ðð ð ð ð ð< */Ø(-ðð à”|ðð    S œ/ðð   $™;ð	ð
  ™+ðð 
ˆuŒ|Ô	ðð ð ð ð ð ð ð r    r>  c                   ó�   ‡ — e Zd Zˆ fd„Z	 	 	 	 	 ddej        deeef         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
ˆ xZS )ÚDonutSwinEncoderc                 ó„  •‡ ‡‡‡— t          ¦   «                              ¦   «          t          ‰j        ¦  «        ‰ _        ‰‰ _        d„ t          j        d‰j        t          ‰j        ¦  «        d¬¦  «        D ¦   «         Št          j        ˆˆˆˆ fd„t          ‰ j        ¦  «        D ¦   «         ¦  «        ‰ _        d‰ _        d S )Nc                 ó6   — g | ]}|                      ¦   «         ‘ŒS r   )Úitem)rž   Úxs     r!   r¡   z-DonutSwinEncoder.__init__.<locals>.<listcomp>§  s    € ÐlÐlÐl˜Aˆq�vŠv‰xŒxÐlÐlÐlr    r   Úcpur  c                 ót  •— g | ]´}t          ‰t          ‰j        d |z  z  ¦  «        ‰d         d |z  z  ‰d         d |z  z  f‰j        |         ‰j        |         ‰t          ‰j        d|…         ¦  «        t          ‰j        d|dz   …         ¦  «        …         |‰j        dz
  k     rt          nd¬¦  «        ‘ŒµS )r*   r   r   N)rQ   r]   r  rF  r´   r  rD  )r>  rz   rG   Údepthsr´   rÎ   Ú
num_layersr’   )rž   Úi_layerrQ   ÚdprrC   rR   s     €€€€r!   r¡   z-DonutSwinEncoder.__init__.<locals>.<listcomp>©  sæ   ø€ ð ð ð ð õ Ø!Ý˜FÔ,¨q°'©zÑ9Ñ:Ô:Ø&/°¤l°q¸'±zÑ&BÀIÈaÄLÐUVÐX_ÑU_ÑD`Ð%aØ œ-¨Ô0Ø$Ô.¨wÔ7Ø!¥# f¤m°H°W°HÔ&=Ñ">Ô">ÅÀVÄ]ÐS`ÐU\Ð_`ÑU`ÐS`ÔEaÑAbÔAbÐ"bÔcØ9@À4Ä?ÐUVÑCVÒ9VÐ9VÕ4Ð4Ð]aðñ ô ðð ð r    F)r>   r?   ÚlenrV  rW  rQ   r   Úlinspacer  rÎ   r   rB  r�   ÚlayersÚgradient_checkpointing)rR   rQ   rC   rY  rT   s   ```@€r!   r?   zDonutSwinEncoder.__init__£  sÍ   øøøøø€ Ý‰Œ×ÒÑÔÐÝ˜fœmÑ,Ô,ˆŒØˆŒØlÐl¥¤°°6Ô3HÍ#ÈfÌmÑJ\ÔJ\ÐejÐ!kÑ!kÔ!kÐlÑlÔlˆÝ”mðð ð ð ð ð ð õ  % T¤_Ñ5Ô5ðñ ô ñ
ô 
ˆŒð ',ˆÔ#Ð#Ð#r    FTr   rš   r¶   NÚoutput_hidden_statesÚ(output_hidden_states_before_downsamplingr*  Úreturn_dictrV   c                 óò  — |rdnd }|rdnd }	|rdnd }
|r?|j         \  }}} |j        |g|¢|‘R Ž }|                     dddd¦  «        }||fz  }|	|fz  }	t          | j        ¦  «        D ]ß\  }} |||||¦  «        }|d         }|d         }|d         }|d         |d         f}|rP|rN|j         \  }}} |j        |g|d         |d         f¢|‘R Ž }|                     dddd¦  «        }||fz  }|	|fz  }	nC|rA|s?|j         \  }}} |j        |g|¢|‘R Ž }|                     dddd¦  «        }||fz  }|	|fz  }	|r|
|dd …         z  }
Œà|st          d„ |||
fD ¦   «         ¦  «        S t          |||
|	¬	¦  «        S )
Nr   r   r   r   r*   r¸   r+   c              3   ó   K  — | ]}|®|V — Œ	d S r×   r   )rž   Úvs     r!   ú	<genexpr>z+DonutSwinEncoder.forward.<locals>.<genexpr>î  s(   è è € ÐmÐm˜qÐ_`Ð_l˜Ð_lÐ_lÐ_lÐ_lÐmÐmr    )r   r   r   r   )r-   r.   ra   rH  r\  r   r   )rR   r   rš   r¶   r^  r_  r*  r`  Úall_hidden_statesÚall_reshaped_hidden_statesÚall_self_attentionsr3   rr   rŠ   Úreshaped_hidden_staterA  rI  r;  rJ  rs   s                       r!   rw   zDonutSwinEncoder.forward¹  s•  € ð #7Ð@˜B˜B¸DÐØ+?Ð%I R RÀTÐ"Ø$5Ð?˜b˜b¸4Ðàð 	CØ)6Ô)<Ñ&ˆJ˜˜;à$6 MÔ$6°zÐ$bÐDTÐ$bÐVaÐ$bÐ$bÐ$bÐ!Ø$9×$AÒ$AÀ!ÀQÈÈ1Ñ$MÔ$MÐ!Ø -Ð!1Ñ1ÐØ&Ð+@Ð*BÑBÐ&å(¨¬Ñ5Ô5ð 	9ð 	9‰OˆAˆ|Ø(˜L¨Ð8HÐJ[Ð]mÑnÔnˆMà)¨!Ô,ˆMØ0=¸aÔ0@Ð-Ø -¨aÔ 0Ðà 1°"Ô 5Ð7HÈÔ7LÐMÐà#ð GÐ(Pð GØ-NÔ-TÑ*�
˜A˜{ð )OÐ(IÔ(NØð)Ø"3°AÔ"6Ð8IÈ!Ô8LÐ!Mð)ØOZð)ð )ð )Ð%ð )>×(EÒ(EÀaÈÈAÈqÑ(QÔ(QÐ%Ø!Ð&GÐ%IÑIÐ!Ø*Ð/DÐ.FÑFÐ*Ð*Ø%ð GÐ.Vð GØ-:Ô-@Ñ*�
˜A˜{à(:¨Ô(:¸:Ð(fÐHXÐ(fÐZeÐ(fÐ(fÐ(fÐ%Ø(=×(EÒ(EÀaÈÈAÈqÑ(QÔ(QÐ%Ø! mÐ%5Ñ5Ð!Ø*Ð/DÐ.FÑFÐ*à ð 9Ø# }°Q°R°RÔ'8Ñ8Ð#øàð 	nÝÐmÐm ]Ð4EÐGZÐ$[ÐmÑmÔmÑmÔmÐmå%Ø+Ø+Ø*Ø#=ð	
ñ 
ô 
ð 	
r    )FFFFT)r   r   r   r?   r   ry   r   rz   r|   r   rw   r}   r~   s   @r!   rO  rO  ¢  sÔ   ø€ € € € € ð,ð ,ð ,ð ,ð ,ð4 */Ø,1Ø@EØ(-Ø#'ð<
ð <
à”|ð<
ð    S œ/ð<
ð   $™;ð	<
ð
 # T™kð<
ð 37¸±+ð<
ð  ™+ð<
ð ˜D‘[ð<
ð 
Ð'Ñ	'ð<
ð <
ð <
ð <
ð <
ð <
ð <
ð <
r    rO  c                   ój   ‡ — e Zd ZU eed<   dZdZdZdZdgZ	 e
j        ¦   «         ˆ fd„¦   «         Zˆ xZS )ÚDonutSwinPreTrainedModelrQ   Údonutrl   )ÚimageTr>  c                 ó´  •— t          ¦   «                              |¦  «         t          |t          ¦  «        rD|j        �t          j        |j        ¦  «         |j        �t          j        |j        ¦  «         dS dS t          |t          ¦  «        rGt          j        |j	        ¦  «         t          j
        |j        |                     ¦   «         ¦  «         dS dS )zInitialize the weightsN)r>   Ú_init_weightsr„   r<   rH   ÚinitÚzeros_rJ   r£   rª   Úcopy_r­   r¬   )rR   ÚmodulerT   s     €r!   rn  z&DonutSwinPreTrainedModel._init_weights  sÑ   ø€ õ 	‰Œ×Ò˜fÑ%Ô%Ð%Ý�fÕ1Ñ2Ô2ð 	`ØÔ Ð,Ý”˜FÔ-Ñ.Ô.Ð.ØÔ)Ð5Ý”˜FÔ6Ñ7Ô7Ð7Ð7Ð7ð 6Ð5å˜Õ 6Ñ7Ô7ð 	`ÝŒK˜Ô;Ñ<Ô<Ð<ÝŒJ�vÔ5°v×7\Ò7\Ñ7^Ô7^Ñ_Ô_Ð_Ð_Ð_ð	`ð 	`r    )r   r   r   r   r   Úbase_model_prefixÚmain_input_nameÚinput_modalitiesÚsupports_gradient_checkpointingÚ_no_split_modulesr   Úno_gradrn  r}   r~   s   @r!   rj  rj  ø  sƒ   ø€ € € € € € ð ÐÐÑØÐØ$€OØ!ÐØ&*Ð#Ø)Ð*Ðà€U„]�_„_ð
`ð 
`ð 
`ð 
`ñ „_ð
`ð 
`ð 
`ð 
`ð 
`r    rj  c                   ó    ‡ — e Zd Zdˆ fd„	Zd„ Ze	 	 	 	 	 	 ddej        dz  dej        dz  de	dz  d	e	dz  d
e	de	dz  de
ez  fd„¦   «         Zˆ xZS )ÚDonutSwinModelTFc                 ó¤  •— t          ¦   «                              |¦  «         || _        t          |j        ¦  «        | _        t          |j        d| j        dz
  z  z  ¦  «        | _        t          ||¬¦  «        | _
        t          || j
        j        ¦  «        | _        |rt          j        d¦  «        nd| _        |                      ¦   «          dS )zû
        add_pooling_layer (bool, *optional*, defaults to `True`):
            Whether to add a pooling layer
        use_mask_token (`bool`, *optional*, defaults to `False`):
            Whether to use a mask token for masked image modeling.
        r*   r   )rS   N)r>   r?   rQ   rZ  rV  rW  rz   rG   Únum_featuresr<   rU   rO  rD   Úencoderr   ÚAdaptiveAvgPool1dÚpoolerÚ	post_init)rR   rQ   Úadd_pooling_layerrS   rT   s       €r!   r?   zDonutSwinModel.__init__  s·   ø€ õ 	‰Œ×Ò˜Ñ Ô Ð ØˆŒÝ˜fœmÑ,Ô,ˆŒÝ Ô 0°1¸¼È1Ñ9LÑ3MÑ MÑNÔNˆÔå-¨fÀ^ÐTÑTÔTˆŒÝ'¨°´Ô0JÑKÔKˆŒà1BÐL•bÔ*¨1Ñ-Ô-Ð-ÈˆŒð 	�ŠÑÔÐÐÐr    c                 ó   — | j         j        S r×   )rU   rA   r  s    r!   Úget_input_embeddingsz#DonutSwinModel.get_input_embeddings&  s   € ØŒÔ/Ð/r    Nrl   rm   r¶   r^  rk   r`  rV   c                 óø  — |�|n| j         j        }|�|n| j         j        }|�|n| j         j        }|€t	          d¦  «        ‚|                      |||¬¦  «        \  }}	|                      ||	|||¬¦  «        }
|
d         }d}| j        �>|                      |                     dd¦  «        ¦  «        }t          j
        |d¦  «        }|s||f|
dd…         z   }|S t          |||
j        |
j        |
j        ¬¦  «        S )	z¿
        bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, num_patches)`):
            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0).
        Nz You have to specify pixel_values)rm   rk   )r¶   r^  r`  r   r   r*   )r   r$   r   r   r   )rQ   r¶   r^  r`  r¦   rU   r}  r  r/   r   r�   r#   r   r   r   )rR   rl   rm   r¶   r^  rk   r`  ÚkwargsÚembedding_outputrš   Úencoder_outputsÚsequence_outputÚpooled_outputrá   s                 r!   rw   zDonutSwinModel.forward)  sN  € ð 2CÐ1NÐ-Ð-ÐTXÔT_ÔTqÐà$8Ð$DÐ Ð È$Ì+ÔJjð 	ð &1Ð%<�k�kÀ$Ä+ÔBYˆàÐÝÐ?Ñ@Ô@Ð@à-1¯_ª_Ø¨/ÐTlð .=ñ .
ô .
Ñ*ÐÐ*ð Ÿ,š,ØØØ/Ø!5Ø#ð 'ñ 
ô 
ˆð *¨!Ô,ˆàˆØŒ;Ð"Ø ŸKšK¨×(AÒ(AÀ!ÀQÑ(GÔ(GÑHÔHˆMÝ!œM¨-¸Ñ;Ô;ˆMàð 	Ø% }Ð5¸ÈÈÈÔ8KÑKˆFàˆMå#Ø-Ø'Ø)Ô7Ø&Ô1Ø#2Ô#Ið
ñ 
ô 
ð 	
r    )TF©NNNNFN)r   r   r   r?   rƒ  r   r   r   r{   r|   r   r#   rw   r}   r~   s   @r!   rz  rz    së   ø€ € € € € ðð ð ð ð ð ð(0ð 0ð 0ð ð 26Ø37Ø)-Ø,0Ø).Ø#'ð5
ð 5
àÔ'¨$Ñ.ð5
ð Ô)¨DÑ0ð5
ð   $™;ð	5
ð
 # T™kð5
ð #'ð5
ð ˜D‘[ð5
ð 
Ð%Ñ	%ð5
ð 5
ð 5
ñ „^ð5
ð 5
ð 5
ð 5
ð 5
r    rz  aì  
    DonutSwin Model transformer with an image classification head on top (a linear layer on top of the final hidden state of
    the [CLS] token) e.g. for ImageNet.

    <Tip>

        Note that it's possible to fine-tune DonutSwin on higher resolution images than the ones it has been trained on, by
        setting `interpolate_pos_encoding` to `True` in the forward of the model. This will interpolate the pre-trained
        position embeddings to the higher resolution.

    </Tip>
    c                   ó˜   ‡ — e Zd Zˆ fd„Ze	 	 	 	 	 	 ddej        dz  dej        dz  dedz  dedz  ded	edz  d
e	e
z  fd„¦   «         Zˆ xZS )ÚDonutSwinForImageClassificationc                 ó@  •— t          ¦   «                              |¦  «         |j        | _        t          |¦  «        | _        |j        dk    r$t          j        | j        j        |j        ¦  «        nt          j        ¦   «         | _	        |  
                    ¦   «          d S r  )r>   r?   Ú
num_labelsrz  rk  r   r—   r|  r  Ú
classifierr€  )rR   rQ   rT   s     €r!   r?   z(DonutSwinForImageClassification.__init__r  s‹   ø€ Ý‰Œ×Ò˜Ñ Ô Ð à Ô+ˆŒÝ# FÑ+Ô+ˆŒ
ð FLÔEVÐYZÒEZÐEZ�BŒI�d”jÔ-¨vÔ/@ÑAÔAÐAÕ`bÔ`kÑ`mÔ`mð 	Œð
 	�ŠÑÔÐÐÐr    NFrl   Úlabelsr¶   r^  rk   r`  rV   c                 óJ  — |�|n| j         j        }|                      |||||¬¦  «        }|d         }	|                      |	¦  «        }
d}|�|                      ||
| j         ¦  «        }|s|
f|dd…         z   }|�|f|z   n|S t          ||
|j        |j        |j        ¬¦  «        S )aŠ  
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the image classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        N)r¶   r^  rk   r`  r   r*   )r'   r(   r   r   r   )	rQ   r`  rk  r�  Úloss_functionr&   r   r   r   )rR   rl   r�  r¶   r^  rk   r`  r…  rÇ   r‰  r(   r'   rá   s                r!   rw   z'DonutSwinForImageClassification.forward€  sá   € ð" &1Ð%<�k�kÀ$Ä+ÔBYˆà—*’*ØØ/Ø!5Ø%=Ø#ð ñ 
ô 
ˆð   œ
ˆà—’ Ñ/Ô/ˆàˆØÐØ×%Ò% f¨f°d´kÑBÔBˆDàð 	FØ�Y ¨¨¨¤Ñ,ˆFØ)-Ð)9�T�G˜fÑ$Ð$¸vÐEå-ØØØ!Ô/ØÔ)Ø#*Ô#Að
ñ 
ô 
ð 	
r    rŠ  )r   r   r   r?   r   r   r   Ú
LongTensorr|   r   r&   rw   r}   r~   s   @r!   rŒ  rŒ  b  s×   ø€ € € € € ð ð ð ð ð ð ð 26Ø*.Ø)-Ø,0Ø).Ø#'ð,
ð ,
àÔ'¨$Ñ.ð,
ð Ô  4Ñ'ð,
ð   $™;ð	,
ð
 # T™kð,
ð #'ð,
ð ˜D‘[ð,
ð 
Ð/Ñ	/ð,
ð ,
ð ,
ñ „^ð,
ð ,
ð ,
ð ,
ð ,
r    rŒ  )rz  rj  rŒ  )1r   Úcollections.abcr…   rº   Údataclassesr   r   r   Ú r   ro  Úactivationsr   Úmodeling_layersr   Úmodeling_utilsr	   Úutilsr
   r   r   r   Úconfiguration_donut_swinr   Ú
get_loggerr   Úloggerr   r#   r&   r8   r:   ÚModuler<   r@   r’   r£   rÕ   rß   ræ   rï   ró   r  r>  rO  rj  rz  rŒ  Ú__all__r   r    r!   ú<module>r      sC  ððð ð
 Ð Ð Ð Ø €€€Ø !Ð !Ð !Ð !Ð !Ð !à €€€Ø Ð Ð Ð Ð Ð à &Ð &Ð &Ð &Ð &Ð &Ø !Ð !Ð !Ð !Ð !Ð !Ø 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø -Ð -Ð -Ð -Ð -Ð -Ø DÐ DÐ DÐ DÐ DÐ DÐ DÐ DÐ DÐ DÐ DÐ DØ 5Ð 5Ð 5Ð 5Ð 5Ð 5ð 
ˆÔ	˜HÑ	%Ô	%€ð €ððñ ô ð
 ðHð Hð Hð Hð H˜[ñ Hô Hñ „ñô ðHð  €ððñ ô ð
 ðHð Hð Hð Hð H˜;ñ Hô Hñ „ñô ðHð& €ððñ ô ð
 ðHð Hð Hð Hð H [ñ Hô Hñ „ñô ðHð,	ð 	ð 	ðð ð ðY-ð Y-ð Y-ð Y-ð Y-˜"œ)ñ Y-ô Y-ð Y-ðz'-ð '-ð '-ð '-ð '-˜rœyñ '-ô '-ð '-ðV%ð %ð %ð %ð %˜BœIñ %ô %ð %ðRZ'ð Z'ð Z'ð Z'ð Z'˜RœYñ Z'ô Z'ð Z'ð|
ð 
ð 
ð 
ð 
˜"œ)ñ 
ô 
ð 
ðð ð ð ð ˜œñ ô ð ð&ð ð ð ð ˜BœIñ ô ð ð 	ð 	ð 	ð 	ð 	�b”iñ 	ô 	ð 	ð%ð %ð %ð %ð %�B”Iñ %ô %ð %ð2uð uð uð uð u�R”Yñ uô uð uðr4ð 4ð 4ð 4ð 4Ð/ñ 4ô 4ð 4ðpS
ð S
ð S
ð S
ð S
�r”yñ S
ô S
ð S
ðl ð`ð `ð `ð `ð `˜ñ `ô `ñ „ð`ð, ðN
ð N
ð N
ð N
ð N
Ð-ñ N
ô N
ñ „ðN
ðb €ððñ ô ð<
ð <
ð <
ð <
ð <
Ð&>ñ <
ô <
ñô ð<
ð~ \Ð
[Ð
[€€€r    