§
    ·¯¸jA‚  ã                   óv  — d Z ddlmZmZmZmZmZmZmZm	Z	 ddl
Z
ddlmZ ddlmc mZ ddlmZmZ ddlmZmZmZmZmZmZmZ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$m%Z%m&Z& dgZ' G d„ dej(        ¦  «        Z) G d„ dej(        ¦  «        Z* G d„ dej(        ¦  «        Z+ G d„ dej(        ¦  «        Z,e! G d„ dej(        ¦  «        ¦   «         Z- G d„ dej(        ¦  «        Z. G d„ dej(        ¦  «        Z/dee0e
j1        f         dej(        dee0e
j1        f         fd„Z2d1de0d edee0ef         fd!„Z3 e& e3d"¬#¦  «         e3d"¬#¦  «         e3d"¬#¦  «         e3¦   «          e3d"¬#¦  «         e3d"¬#¦  «         e3d"¬#¦  «         e3d"¬#¦  «        d$œ¦  «        Z4d2d&e0d'e5d ede/fd(„Z6e%d2d'e5d ede/fd)„¦   «         Z7e%d2d'e5d ede/fd*„¦   «         Z8e%d2d'e5d ede/fd+„¦   «         Z9e%d2d'e5d ede/fd,„¦   «         Z:e%d2d'e5d ede/fd-„¦   «         Z;e%d2d'e5d ede/fd.„¦   «         Z<e%d2d'e5d ede/fd/„¦   «         Z=e%d2d'e5d ede/fd0„¦   «         Z>dS )3a  LowFormer
LowFormer: Hardware Efficient Design for Convolutional Transformer Backbones (WACV 2025)
- paper: https://arxiv.org/abs/2409.03460
- code: https://github.com/altair199797/LowFormer
@article{Nottebaum2024LowFormerHE,
  title={LowFormer: Hardware Efficient Design for Convolutional Transformer Backbones},
  author={Moritz Nottebaum and Matteo Dunnhofer and Christian Micheloni},
  journal={2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2024},
  pages={7008-7018},
}

The LowFormer-E1/E2/E3 edge GPU variants are introduced in the journal extension
Beyond MACs: Hardware Efficient Architecture Design for Vision Backbones (IJCV 2026)
- paper: https://arxiv.org/abs/2603.26551
- code: https://github.com/altair199797/LowFormer
@article{Nottebaum2026BeyondMACs,
  author  = {Nottebaum, Moritz and Dunnhofer, Matteo and Micheloni, Christian},
  title   = {Beyond {MAC}s: Hardware Efficient Architecture Design for Vision Backbones},
  journal = {International Journal of Computer Vision},
  year    = {2026},
  volume  = {134},
  number  = {6},
  pages   = {295},
  doi     = {10.1007/s11263-026-02873-5},
  url     = {https://doi.org/10.1007/s11263-026-02873-5}
}

Modifications by / Copyright 2026 Ryan Hou & Ross Wightman, original copyrights below
é    )ÚAnyÚDictÚListÚOptionalÚSetÚTupleÚTypeÚUnionN©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚDropPathÚ
GroupNorm1ÚLinearÚSelectAdaptivePool2dÚcalculate_drop_path_ratesÚget_device_dtypeÚtrunc_normal_Úuse_fused_attné   )Úbuild_model_with_cfg)Úfeature_take_indices)Úregister_notrace_module)Úcheckpoint_seq)Úregister_modelÚgenerate_default_cfgsÚ	LowFormerc                   óv   ‡ — e Zd Z	 	 d
dej        deej                 defˆ fd„Zdej	        dej	        fd	„Z
ˆ xZS )ÚResidualBlockNç        ÚmainÚshortcutÚ	drop_pathc                 ó¾   •— t          ¦   «                              ¦   «          || _        || _        |dk    rt	          |¦  «        nt          j        ¦   «         | _        d S ©Nr    )ÚsuperÚ__init__r!   r"   r   ÚnnÚIdentityr#   )Úselfr!   r"   r#   Ú	__class__s       €úe/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/lowformer.pyr'   zResidualBlock.__init__:   sO   ø€ õ 	‰Œ×ÒÑÔÐØˆŒ	Ø ˆŒØ09¸B²°� )Ñ,Ô,Ð,ÅBÄKÁMÄMˆŒˆˆó    ÚxÚreturnc                 ó¶   — | j         €|                      |¦  «        S |                      |                      |¦  «        ¦  «        |                       |¦  «        z   S ©N)r"   r!   r#   ©r*   r.   s     r,   ÚforwardzResidualBlock.forwardE   sG   € ØŒ=Ð Ø—9’9˜Q‘<”<ÐØ�~Š~˜dŸiši¨™lœlÑ+Ô+¨d¯mªm¸AÑ.>Ô.>Ñ>Ð>r-   r%   )Ú__name__Ú
__module__Ú__qualname__r(   ÚModuler   Úfloatr'   ÚtorchÚTensorr3   Ú__classcell__©r+   s   @r,   r   r   9   sŸ   ø€ € € € € ð -1Ø!ð		Rð 	Rà”)ð	Rð ˜rœyÔ)ð	Rð ð		Rð 	Rð 	Rð 	Rð 	Rð 	Rð?˜œð ?¨%¬,ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?r-   r   c                   óÒ   ‡ — e Zd ZdZddddej        ej        ddfdededed	ed
edede	e
ej                          de	e
ej                          fˆ fd„Zdej        dej        fd„Zˆ xZS )Ú	ConvLayeru)  Conv + optional norm + optional act.

    NOTE deliberately not timm's ConvNormAct: LowFormer needs activation-without-norm
    (the MBConv depthwise convs), which ConvNormAct cannot express â€” its norm and act are
    combined via get_norm_act_layer, which returns None for norm_layer=None.
    é   r   FNÚin_channelsÚout_channelsÚkernel_sizeÚstrideÚgroupsÚuse_biasÚ
norm_layerÚ	act_layerc           	      ó&  •— t          ¦   «                              ¦   «          |	|
dœ}|dz  }t          j        ||f|||||dœ|¤Ž| _        |�	 ||fi |¤Žnt          j        ¦   «         | _        |�
 |¦   «         nt          j        ¦   «         | _        d S )N©ÚdeviceÚdtypeé   )rB   rC   ÚpaddingrD   Úbias)r&   r'   r(   ÚConv2dÚconvr)   ÚnormÚact)r*   r@   rA   rB   rC   rD   rE   rF   rG   rJ   rK   ÚddrM   r+   s                €r,   r'   zConvLayer.__init__S   sÀ   ø€ õ 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ Ñ"ˆÝ”IØØð	
ð $ØØØØð	
ð 	
ð ð	
ð 	
ˆŒ	ð 7AÐ6L�J�J˜|Ð2Ð2¨rÐ2Ð2Ð2ÕRTÔR]ÑR_ÔR_ˆŒ	Ø"+Ð"7�9�9‘;”;�;½R¼[¹]¼]ˆŒˆˆr-   r.   r/   c                 ó„   — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|S r1   )rP   rQ   rR   r2   s     r,   r3   zConvLayer.forwardp   s4   € Ø�IŠI�a‰LŒLˆØ�IŠI�a‰LŒLˆØ�HŠH�Q‰KŒKˆØˆr-   )r4   r5   r6   Ú__doc__r(   ÚBatchNorm2dÚReLUÚintÚboolr   r	   r7   r'   r9   r:   r3   r;   r<   s   @r,   r>   r>   K   s  ø€ € € € € ðð ð  !ØØØ"Ø46´NØ35´7ØØðKð KàðKð ðKð ð	Kð
 ðKð ðKð ðKð !  b¤i¤Ô1ðKð    R¤Y¤Ô0ðKð Kð Kð Kð Kð Kð:˜œð ¨%¬,ð ð ð ð ð ð ð ð r-   r>   c                   óæ   ‡ — e Zd Zddddddej        ej        ddf
dededed	ed
ee         dedede	dee
ej                          dee
ej                          fˆ fd„Zdej        dej        fd„Zˆ xZS )ÚMBConvr?   r   Né   Fr@   rA   rB   rC   Úmid_channelsÚexpand_ratioÚexpand_groupsrE   rF   rG   c           	      ó  •— t          ¦   «                              ¦   «          ||dœ}|pt          ||z  ¦  «        }t          ||dfd||d |
dœ|¤Ž| _        t          |||f|||d |
dœ|¤Ž| _        t          ||df|d|	d dœ|¤Ž| _        d S )NrI   r   )rC   rD   rE   rF   rG   F©rD   rE   rF   rG   )r&   r'   Úroundr>   Úinverted_convÚ
depth_convÚ
point_conv©r*   r@   rA   rB   rC   r]   r^   r_   rE   rF   rG   rJ   rK   rS   r+   s                 €r,   r'   zMBConv.__init__x   s  ø€ õ 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ#ÐH¥u¨[¸<Ñ-GÑ'HÔ'Hˆõ 'ØØØð

ð Ø ØØØð

ð 

ð ð

ð 

ˆÔõ $ØØØð

ð ØØØØð

ð 

ð ð

ð 

ˆŒõ $ØØØð	
ð !ØØ!Øð	
ð 	
ð ð	
ð 	
ˆŒˆˆr-   r.   r/   c                 ó„   — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|S r1   )rc   rd   re   r2   s     r,   r3   zMBConv.forward¯   s<   € Ø×Ò˜qÑ!Ô!ˆØ�OŠO˜AÑÔˆØ�OŠO˜AÑÔˆØˆr-   ©r4   r5   r6   r(   rV   ÚReLU6rX   r   r8   rY   r	   r7   r'   r9   r:   r3   r;   r<   s   @r,   r[   r[   w   s	  ø€ € € € € ð
  !ØØ*.Ø"#Ø!"Ø"Ø46´NØ35´8ØØð5
ð 5
àð5
ð ð5
ð ð	5
ð
 ð5
ð # 3œ-ð5
ð  ð5
ð ð5
ð ð5
ð !  b¤i¤Ô1ð5
ð    R¤Y¤Ô0ð5
ð 5
ð 5
ð 5
ð 5
ð 5
ðn˜œð ¨%¬,ð ð ð ð ð ð ð ð r-   r[   c                   óæ   ‡ — e Zd Zddddddej        ej        ddf
dededed	ed
ee         dedede	dee
ej                          dee
ej                          fˆ fd„Zdej        dej        fd„Zˆ xZS )ÚFusedMBConvr?   r   Nr\   Fr@   rA   rB   rC   r]   r^   r_   rE   rF   rG   c                 óà   •— t          ¦   «                              ¦   «          ||dœ}|pt          ||z  ¦  «        }t          ||||f|||	|
dœ|¤Ž| _        t          ||dfdd|	d dœ|¤Ž| _        d S )NrI   ra   r   F)r&   r'   rb   r>   Úspatial_convre   rf   s                 €r,   r'   zFusedMBConv.__init__·   sÁ   ø€ õ 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ#ÐH¥u¨[¸<Ñ-GÑ'HÔ'Hˆå%ØØØØð	

ð
 !ØØ!Øð

ð 

ð ð

ð 

ˆÔõ $ØØØð	
ð ØØ!Øð	
ð 	
ð ð	
ð 	
ˆŒˆˆr-   r.   r/   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r1   )rm   re   r2   s     r,   r3   zFusedMBConv.forwardà   s+   € Ø×Ò˜aÑ Ô ˆØ�OŠO˜AÑÔˆØˆr-   rh   r<   s   @r,   rk   rk   ¶   s	  ø€ € € € € ð
  !ØØ*.Ø"#Ø!"Ø"Ø46´NØ35´8ØØð'
ð '
àð'
ð ð'
ð ð	'
ð
 ð'
ð # 3œ-ð'
ð  ð'
ð ð'
ð ð'
ð !  b¤i¤Ô1ð'
ð    R¤Y¤Ô0ð'
ð '
ð '
ð '
ð '
ð '
ðR˜œð ¨%¬,ð ð ð ð ð ð ð ð r-   rk   c                   ó–   ‡ — e Zd ZU dZej        j        e         ed<   	 	 	 	 	 	 dde	d	e
d
e	de	def
ˆ fd„Zdej        dej        fd„Zˆ xZS )ÚConvAttentionzVLow-frequency / conv-projected attention with strided downsample and learned upsample.Ú
fused_attnç      à?é   é   FNÚ	input_dimÚhead_dim_mulÚ
att_strideÚ
att_kernelÚfuse_out_projc           	      óÚ  •— t          ¦   «                              ¦   «          ||dœ}t          t          d||z  dz  ¦  «        ¦  «        | _        t          || j        z  |z  ¦  «        | _        d| _        | j        dz  | _        || _        t          ¦   «         | _
        t          | j        | j        z  | j        z  ¦  «        }	t          ||f|||t          j        d dœ|¤Ž| _        t          j        ||	fddddd	œ|¤Ž| _        | j        | j        z  | _        |rjt          j        ¦   «         | _        |dk    r#t          j        | j        |fdddd
œ|¤Ž| _        d S t          j        | j        |f|dz  ||dz  d
œ|¤Ž| _        d S t          j        | j        |fdddd
œ|¤Ž| _        |dk    rt          j        ||fddd|dœ|¤Ž| _        d S t          j        ||f|dz  ||dz  |dœ|¤Ž| _        d S )NrI   r   é   r?   g      à¿)rB   rC   rD   rF   rG   r   F)rB   rC   rM   rN   )rB   rC   rM   rL   )rB   rC   rM   rD   )r&   r'   rX   ÚmaxÚ	num_headsÚhead_dimÚnum_keysÚscalerw   r   rq   r>   r(   rV   Ú	conv_projrO   ÚpwiseÚo_proj_inpdimr)   Úo_projÚConvTranspose2dÚ
upsampling)r*   ru   rv   rw   rx   ry   rJ   rK   rS   Ú	total_dimr+   s             €r,   r'   zConvAttention.__init__î   s‹  ø€ õ 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆÝ�S  Y°Ñ%=À"Ñ$DÑEÔEÑFÔFˆŒÝ˜Y¨$¬.Ñ8¸LÑHÑIÔIˆŒØˆŒØ”] dÑ*ˆŒ
Ø$ˆŒÝ(Ñ*Ô*ˆŒå˜œ¨¬Ñ6¸¼ÑFÑGÔGˆ	å"ØØð	
ð #ØØÝ”~Øð	
ð 	
ð ð	
ð 	
ˆŒõ ”Y˜y¨)ÐjÀÈ1ÐVWÐ^cÐjÐjÐgiÐjÐjˆŒ
à!œ]¨T¬^Ñ;ˆÔð ð 	tÝœ+™-œ-ˆDŒKØ˜QŠˆÝ"$Ô"4ØÔ&¨	ð#]Ø?@ÈÐTUð#]ð #]ØY[ð#]ð #]�”��õ #%Ô"4ØÔ&¨	ð#bà *¨Q¡°zÈ:ÐYZÉ?ð#bð #bà^`ð#bð #b�”��õ œ) DÔ$6¸	ÐlÈqÐYZÐdeÐlÐlÐikÐlÐlˆDŒKØ˜QŠˆÝ"$Ô"4Ø˜yð#fØ67ÀÈ1ÐU^ð#fð #fØbdð#fð #f�”��õ #%Ô"4Ø˜yð#tà *¨Q¡°zÈ:ÐYZÉ?Ðclð#tð #tàprð#tð #t�”��r-   r.   r/   c                 óø  — |j         dd …         \  }}|                      |¦  «        }|                      |¦  «        }|                     ¦   «         \  }}}}|                     || j        | j        | j        z  ||z  ¦  «        }	|	                     dddd¦  «        }	|	 	                    dd¬¦  «        \  }
}}| j
        rt          j        |
||¦  «        }n>|
| j        z  }
|
|                     dd¦  «        z  }|                     d¬¦  «        }||z  }|                      |                     dddd¦  «                             || j        ||¦  «        ¦  «        }|                      |¦  «        }|dd |…d |…f         S )	Néþÿÿÿr   r   r?   rL   )Údiméÿÿÿÿ.)Úshaper�   r‚   ÚsizeÚreshaper}   r   r~   ÚpermuteÚchunkrq   ÚFÚscaled_dot_product_attentionr€   Ú	transposeÚsoftmaxr„   rƒ   r†   )r*   r.   ÚHÚWÚxoutÚNÚ_ÚhÚwÚqkvÚqÚkÚvÚvaluesÚattnÚos                   r,   r3   zConvAttention.forward%  si  € ØŒw�r�s�sŒ|‰ˆˆ1à�~Š~˜aÑ Ô ˆØ�zŠz˜$ÑÔˆà—Y’Y‘[”[‰
ˆˆ1ˆa�Ø�lŠl˜1˜dœn¨d¬m¸d¼mÑ.KÈQÐQRÉUÑSÔSˆØ�kŠk˜!˜Q  1Ñ%Ô%ˆØ—)’)˜A 1�)Ñ%Ô%‰ˆˆ1ˆaàŒ?ð 	ÝÔ3°A°q¸!Ñ<Ô<ˆFˆFà�D”J‘ˆAØ�q—{’{ 2 rÑ*Ô*Ñ*ˆDØ—<’< B�<Ñ'Ô'ˆDØ˜A‘XˆFØ�KŠK˜Ÿš q¨!¨Q°Ñ2Ô2×:Ò:¸1¸dÔ>PÐRSÐUVÑWÔWÑXÔXˆà�OŠO˜AÑÔˆà��b�q�b˜"˜1˜"�Œ~Ðr-   )rr   rs   rt   FNN)r4   r5   r6   rU   r9   ÚjitÚFinalrY   Ú__annotations__rX   r8   r'   r:   r3   r;   r<   s   @r,   rp   rp   è   sÖ   ø€ € € € € € à`Ð`à”	” Ô%Ð%Ð%Ñ%ð
 #&ØØØ"'ØØð5tð 5tàð5tð  ð5tð ð	5tð
 ð5tð  ð5tð 5tð 5tð 5tð 5tð 5tðn˜œð ¨%¬,ð ð ð ð ð ð ð ð r-   rp   c                   óÔ   ‡ — e Zd ZdZdej        ej        ddddddddddfded	ed
e	ej
                 de	ej
                 dedededededededefˆ fd„Zdej        dej        fd„Zˆ xZS )ÚLowFormerBlockzbAttention (context) and MLP branches followed by a local conv branch, all with identity residuals.rs   Fr   Tr    Nr@   r^   rF   rG   Ú
fused_convr_   r¡   Úattn_mlpÚattn_mlp_ratiorw   Ú	proj_dropr#   c                 óP  •— t          ¦   «                              ¦   «          ||dœ}|
dk    rdnd}|r\t          d||
|d|dœ|¤Ž}|r t          j        t          |fi |¤Ž|¦  «        }t          |t          j        ¦   «         |¦  «        | _        nt          j        ¦   «         | _        |r˜t          t          j        t          |fi |¤Žt          j	        |||	z  fddi|¤Žt          j
        ¦   «         t          j	        ||	z  |fddi|¤Žt          j        |¦  «        ¦  «        t          j        ¦   «         |¦  «        | _        nt          j        ¦   «         | _        |r|dk     rt          nt          } |d||||d	||d
œ|¤Ž}t          |t          j        ¦   «         |¦  «        | _        d S )NrI   r   é   r?   rr   )ru   rw   rx   rv   ry   rB   é   T)r@   rA   r^   r_   rE   rF   rG   © )r&   r'   rp   r(   Ú
Sequentialr   r   r)   r¡   rO   ÚGELUÚDropoutÚmlprk   r[   Úlocal)r*   r@   r^   rF   rG   r¨   r_   r¡   r©   rª   rw   r«   r#   rJ   rK   rS   rx   Úattn_moduleÚ	block_clsÚlocal_moduler+   s                       €r,   r'   zLowFormerBlock.__init__A  sî  ø€ õ" 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ$ qš.˜.�Q�Q¨aˆ
àð 	&Ý%2ð &Ø%Ø%Ø%Ø Ø(ð&ð &ð ð&ð &ˆKð ð XÝ œm­J°{Ð,IÐ,IÀbÐ,IÐ,IÈ;ÑWÔW�Ý% kµ2´;±=´=À)ÑLÔLˆDŒIˆIåœ™œˆDŒIàð 	%Ý$Ý”Ý˜{Ð1Ð1¨bÐ1Ð1Ý”I˜k¨;¸Ñ+GÐ]Ð]ÐUVÐ]ÐZ\Ð]Ð]Ý”G‘I”IÝ”I˜k¨NÑ:¸KÐ]Ð]ÐUVÐ]ÐZ\Ð]Ð]Ý”J˜yÑ)Ô)ñô õ ”‘”Øñ
ô 
ˆDŒHˆHõ ”{‘}”}ˆDŒHà#-ÐO°+ÀÒ2CÐ2C•K�KÍˆ	Ø �yð 	
Ø#Ø$Ø%Ø'ØØ!Øð	
ð 	
ð ð	
ð 	
ˆõ # <µ´±´À	ÑJÔJˆŒ
ˆ
ˆ
r-   r.   r/   c                 ó„   — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|S r1   )r¡   r³   r´   r2   s     r,   r3   zLowFormerBlock.forward�  s4   € Ø�IŠI�a‰LŒLˆØ�HŠH�Q‰KŒKˆØ�JŠJ�q‰MŒMˆØˆr-   )r4   r5   r6   rU   r(   rV   Ú	HardswishrX   r8   r	   r7   rY   r'   r9   r:   r3   r;   r<   s   @r,   r§   r§   >  s1  ø€ € € € € ØlÐlð
 #$Ø*,¬.Ø)+¬Ø$Ø!"ØØ!Ø"#ØØ!Ø!ØØð>Kð >Kàð>Kð  ð>Kð ˜RœYœð	>Kð
 ˜BœI”ð>Kð ð>Kð ð>Kð ð>Kð ð>Kð  ð>Kð ð>Kð ð>Kð ð>Kð >Kð >Kð >Kð >Kð >Kð@˜œð ¨%¬,ð ð ð ð ð ð ð ð r-   r§   c            (       ó¤  ‡ — e Zd Zddddddddej        ej        dddddd	d
d	d	fdee         dee         dededede	eef         de
de
de
de
deej                 deej                 dedededede
dee	e
df                  def&ˆ fd„Zd ej        d!d	fd"„Zej        j        d!efd#„¦   «         Zej        j        d:d%ed!eeef         fd&„¦   «         Zej        j        d;d'efd(„¦   «         Zej        j        d!ej        fd)„¦   «         Zd<dedee         fd*„Z	 	 	 	 	 d=d,ej        d-eeeee         f                  d.ed/ed0ed1ed!eeej                 e	ej        eej                 f         f         fd2„Z	 	 	 d>d-eeee         f         d3ed4ed!ee         fd5„Zd,ej        d!ej        fd6„Z d:d,ej        d7ed!ej        fd8„Z!d,ej        d!ej        fd9„Z"ˆ xZ#S )?r   r?   éè  Úavg)i   i@  r    rs   TrL   Nr   Ú
width_listÚ
depth_listÚin_chansÚnum_classesÚglobal_poolÚhead_widthsÚ	drop_rateÚproj_drop_rateÚdrop_path_rater^   rF   rG   r¨   r¡   r©   rª   Ústem_expand_ratioÚdownsample_expand_ratios.r_   c                 ó¢  •— t          ¦   «                              ¦   «          ||dœ}|| _        || _        || _        g | _        d| _        t          |¦  «        dz
  }|€|
f|z  }t          |¦  «        |k    sJ ‚t          |¦  «        }t          |	t          |¦  «        ¦  «        }|rt          nt          }t          d||d         dd||dœ|¤Žg}|d         }t          |d         ¦  «        D ]N} |d||d||d||d	œ|¤Ž}|                     t!          |t#          j        ¦   «         ||         ¦  «        ¦  «         ŒOt#          j        |Ž | _        g }d}|d         } t+          t-          |dd…         |dd…         ¦  «        d¬
¦  «        D �]’\  }!\  }"}#|| | |#z   …         }$| |#z  } g }%|!dk    rŒ |d||"d||!dz
           |d||d	œ|¤Ž}&|%                     t!          |&d¦  «        ¦  «         |"}t          |#¦  «        D ]<}|%                     t/          d||
||||||||!dk    rdnd||$|         dœ|¤Ž¦  «         Œ=n‰|#dk    rƒt          |#¦  «        D ]s}|dk    rdnd}' |d||"|'|'dk    r||!dz
           n|
|d||d	œ|¤Ž}|%                     t!          ||'dk    rt#          j        ¦   «         nd|$|         ¦  «        ¦  «         |"}Œt|                     t#          j        |%Ž ¦  «         ||%rdndz  }| j                             t1          ||d|!dz
  › �¬¦  «        ¦  «         �Œ”t#          j        |Ž | _        |d         | _        |d         | _        t9          |¬¦  «        | _        |rt#          j        d¦  «        nt#          j        ¦   «         | _        t          d|d         |d         d||dœ|¤Ž| _         t#          j        tC          |d         |d         fddi|¤Žt#          j"        |d         fi |¤Ž |¦   «         ¦  «        | _#        |dk    rtC          |d         |fi |¤Žnt#          j        ¦   «         | _$        |  %                    | j&        ¦  «         dS )u	  LowFormer backbone with classification head.

        Args:
            width_list: Channel width of the stem and each of the four stages.
            depth_list: Number of blocks in the stem and each of the four stages.
            in_chans: Number of input image channels.
            num_classes: Number of classifier output classes.
            global_pool: Global pooling type, either 'avg' or '' to disable pooling.
            head_widths: Hidden widths of the classification head (in_conv, pre_classifier).
            drop_rate: Classifier dropout rate.
            proj_drop_rate: Dropout rate at the end of the attention-stage (LowFormerBlock) MLPs.
            drop_path_rate: Stochastic depth rate.
            expand_ratio: Default expand ratio of MBConv / FusedMBConv blocks.
            norm_layer: Normalization layer.
            act_layer: Activation layer.
            fused_conv: Use FusedMBConv instead of MBConv in the stem, conv stages, attention
                stage downsamples and block local branches; also folds the attention output
                projection into its upsampling layer.
            attn: Include the attention (ConvAttention) branch in LowFormerBlocks.
            attn_mlp: Include the MLP branch in LowFormerBlocks.
            attn_mlp_ratio: MLP expansion ratio in the attention (LowFormerBlock) stages.
            stem_expand_ratio: Expand ratio of the stem blocks.
            downsample_expand_ratios: Per-stage expand ratios of the four stride-2 downsample
                blocks, None uses `expand_ratio` everywhere.
            expand_groups: Groups of the expand conv in MBConv / FusedMBConv blocks.

        The ablation flags of the original implementation map onto these args as follows:
        `fastit=True` â†’ `fused_conv=True, attn_mlp_ratio=4`; `bigit=True` â†’
        `downsample_expand_ratios=(6, 6, 6, 6)`; `huge_model=True` â†’ `stem_expand_ratio=4`;
        `grouping` â†’ `expand_groups`; `noattention=True` â†’ `attn=False`; `mlpremoved=True` â†’
        `attn_mlp=False`.
        rI   Fr   Nr   r?   rL   )r@   rA   rB   rC   rF   rG   )r@   rA   rC   r^   r_   rE   rF   rG   )Ústart)r@   r^   rF   rG   r¨   r_   r¡   r©   rª   rw   r«   r#   ústages.)Únum_chsÚ	reductionÚmoduler‹   ©Ú	pool_type)r@   rA   rB   rF   rG   rN   r¯   )'r&   r'   rÀ   r¿   rÃ   Úfeature_infoÚgrad_checkpointingÚlenÚtupler   Úsumrk   r[   r>   ÚrangeÚappendr   r(   r)   r°   ÚstemÚ	enumerateÚzipr§   ÚdictÚstagesÚnum_featuresÚhead_hidden_sizer   rÁ   ÚFlattenÚflattenÚin_convr   Ú	LayerNormÚpre_classifierÚ
classifierÚapplyÚ_init_weights))r*   r½   r¾   r¿   rÀ   rÁ   rÂ   rÃ   rÄ   rÅ   r^   rF   rG   r¨   r¡   r©   rª   rÆ   rÇ   r_   rJ   rK   rS   Ú
num_stagesÚdprr¶   Ústem_blocksr@   ÚiÚblockrÛ   rÌ   Ú	block_idxÚ	stage_numÚwidthÚdepthÚ	stage_dprÚblocksÚ
downsamplerC   r+   s)                                           €r,   r'   zLowFormer.__init__‰  s™  ø€ õp 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ&ˆÔØ ˆŒØ"ˆŒØˆÔØ"'ˆÔå˜‘_”_ qÑ(ˆ
Ø#Ð+Ø(4 ¸Ñ'CÐ$ÝÐ+Ñ,Ô,°
Ò:Ð:Ð:Ð:Ý#(Ð)AÑ#BÔ#BÐ õ
 (¨½¸J¹¼ÑHÔHˆØ#-Ð9•K�Kµ6ˆ	õ ð Ø$Ø'¨œ]ØØØ%Ø#ðð ð ðð ð

ˆð ! ”mˆÝ�z !”}Ñ%Ô%ð 	Lð 	LˆAØ�Ið 
Ø'Ø(ØØ.Ø+ØØ%Ø#ð
ð 
ð ð
ð 
ˆEð ×Ò�}¨UµB´K±M´MÀ3ÀqÄ6ÑJÔJÑKÔKÐKÐKÝ”M ;Ð/ˆŒ	ð ˆØˆ	Ø˜q”Mˆ	Ý)2µ3°zÀ!À"À"´~ÀzÐRSÐRTÐRTÄ~Ñ3VÔ3VÐ^_Ð)`Ñ)`Ô)`ð :	bñ :	bÑ%ˆI‘~˜˜uØ˜I i°%Ñ&7Ð7Ô8ˆIØ˜ÑˆIØˆFØ˜AŠ~ˆ~Ø&˜Yð 
Ø +Ø!&ØØ!9¸)Àa¹-Ô!HØ"/Ø"Ø)Ø'ð
ð 
ð ð
ð 
�
ð —’�m¨J¸Ñ=Ô=Ñ>Ô>Ð>Ø#�Ý˜u™œð ð �AØ—M’MÝ&ð Ø(3Ø)5Ø'1Ø&/Ø'1Ø*7Ø!%Ø%-Ø+9Ø,5¸ªN¨N q qÀØ&4Ø&/°¤lðð ð !ðð ñô ð ð ðð$ ˜’�Ý˜u™œð (ð (�AØ"# q¢& &˜Q˜Q¨a�FØ%˜Ið Ø$/Ø%*Ø%àGMÐQRÂ{À{Ð4°YÀ±]ÔCÐCÐXdà&3Ø!&Ø#-Ø"+ðð ð ðð �Eð —M’M¥-°ÈÐRSÊÈµr´{±}´}°}ÐY]Ð_hÐijÔ_kÑ"lÔ"lÑmÔmÐmØ"'�K�Kà�MŠM�"œ-¨Ð0Ñ1Ô1Ð1Ø˜fÐ+˜˜¨!Ñ+ˆIØÔ×$Ò$Ý˜[°IÐF_ÐPYÐ\]ÑP]ÐF_ÐF_Ð`Ñ`Ô`ñbô bð bñ bå”m VÐ,ˆŒð ' rœNˆÔØ +¨B¤ˆÔÝ/¸+ÐFÑFÔFˆÔØ(3ÐF•r”z !‘}”}�}½¼¹¼ˆŒÝ ð 
Ø" 2œØ$ QœØØ!Øð
ð 
ð ð
ð 
ˆŒõ !œmÝ�;˜q”> ;¨q¤>ÐDÐD¸ÐDÀÐDÐDÝŒL˜ QœÐ.Ð.¨2Ð.Ð.ØˆI‰KŒKñ
ô 
ˆÔð
 HSÐUVÂÀ�& ¨Q¤°ÐCÐCÀÐCÐCÐCÕ\^Ô\gÑ\iÔ\iˆŒà�
Š
�4Ô%Ñ&Ô&Ð&Ð&Ð&r-   rÍ   r/   c                 óÆ   — t          |t          j        ¦  «        rDt          |j        d¬¦  «         |j        �)t          j                             |j        d¦  «         d S d S d S )Ng{®Gáz”?)Ústdr   )Ú
isinstancer(   r   r   ÚweightrN   ÚinitÚ	constant_)r*   rÍ   s     r,   rå   zLowFormer._init_weightsH  sc   € Ý�f�bœiÑ(Ô(ð 	2Ý˜&œ-¨TÐ2Ñ2Ô2Ð2ØŒ{Ð&Ý”×!Ò! &¤+¨qÑ1Ô1Ð1Ð1Ð1ð	2ð 	2à&Ð&r-   c                 ó   — t          ¦   «         S r1   )Úset©r*   s    r,   Úno_weight_decayzLowFormer.no_weight_decayN  s   € å‰uŒuˆr-   FÚcoarsec                 ó6   — t          d|rdndd fg¬¦  «        }|S )Nz
^stem\.\d+z^stages\.(\d+)z^stages\.(\d+)\.(\d+))r×   rð   )rÚ   )r*   rü   Úmatchers      r,   Úgroup_matcherzLowFormer.group_matcherR  s9   € åØà&,ÐJÐ"Ð"Ð2JÈDÐQðð
ñ 
ô 
ˆð ˆr-   Úenablec                 ó   — || _         d S r1   )rÑ   )r*   r   s     r,   Úset_grad_checkpointingz LowFormer.set_grad_checkpointing\  s   € à"(ˆÔÐÐr-   c                 ó   — | j         S r1   )rã   rú   s    r,   Úget_classifierzLowFormer.get_classifier`  s
   € àŒÐr-   c                 ó”  — t          | ¦  «        }| j        }|| _        |�]t          |¬¦  «        | _        |rt          j        d¦  «        nt          j        ¦   «         | _        | j         	                    |¦  «         |dk    rt          | j        |fi |¤Žnt          j        ¦   «         | _        | j         	                    |¦  «         d S )NrÎ   r   r   )r   ÚtrainingrÀ   r   rÁ   r(   rÞ   r)   rß   Útrainr   rÝ   rã   )r*   rÀ   rÁ   rS   Úwas_trainings        r,   Úreset_classifierzLowFormer.reset_classifierd  sÎ   € Ý˜dÑ#Ô#ˆØ”}ˆØ&ˆÔØÐ"Ý3¸kÐJÑJÔJˆDÔØ,7ÐJ�2œ: a™=œ=˜=½R¼[¹]¼]ˆDŒLØÔ×"Ò" <Ñ0Ô0Ð0ð ˜1Š_ˆ_õ !ØÔ! ;ð
ð 
Ø24ð
ð 
ð 
å"$¤+¡-¤-ð 	Œð 	Œ×Ò˜lÑ+Ô+Ð+Ð+Ð+r-   ÚNCHWr.   ÚindicesrQ   Ú
stop_earlyÚ
output_fmtÚintermediates_onlyc                 óø  — |dv s
J d¦   «         ‚g }t          t          | j        ¦  «        |¦  «        \  }}	|                      |¦  «        }t          j                             ¦   «         s|s| j        }
n| j        d|	dz   …         }
t          |
¦  «        D ]_\  }}| j        r/t          j                             ¦   «         st          ||¦  «        }n ||¦  «        }||v r| 
                    |¦  «         Œ`|r|S ||fS )ap  Forward features that returns intermediates.

        Args:
            x: Input image tensor.
            indices: Take last n blocks if int, all if None, select matching indices if sequence.
            norm: Apply norm layer to compatible intermediates (no-op, LowFormer has no final norm).
            stop_early: Stop iterating over blocks when last desired intermediate hit.
            output_fmt: Shape of intermediate feature outputs.
            intermediates_only: Only return intermediate features.

        Returns:
            List of intermediate features or tuple of (final features, intermediates).
        )r
  zOutput shape must be NCHW.Nr   )r   rÒ   rÛ   r×   r9   r£   Úis_scriptingrØ   rÑ   r   rÖ   )r*   r.   r  rQ   r  r  r  ÚintermediatesÚtake_indicesÚ	max_indexrÛ   Úfeat_idxÚstages                r,   Úforward_intermediateszLowFormer.forward_intermediatesq  s  € ð, ˜YÐ&Ð&Ð&Ð(DÑ&Ô&Ð&ØˆÝ"6µs¸4¼;Ñ7GÔ7GÈÑ"QÔ"QÑˆ�ià�IŠI�a‰LŒLˆÝŒ9×!Ò!Ñ#Ô#ð 	1¨:ð 	1Ø”[ˆFˆFà”[  )¨a¡- Ô0ˆFÝ(¨Ñ0Ô0ð 	(ð 	(‰OˆH�eØÔ&ð ­u¬y×/EÒ/EÑ/GÔ/Gð Ý" 5¨!Ñ,Ô,��à�E˜!‘H”H�Ø˜<Ð'Ð'Ø×$Ò$ QÑ'Ô'Ð'øàð 	!Ø Ð à�-ÐÐr-   Ú
prune_normÚ
prune_headc                 ó¬   — t          t          | j        ¦  «        |¦  «        \  }}| j        d|dz   …         | _        |r|                      d¦  «         |S )aE  Prune layers not required for specified intermediates.

        Args:
            indices: Indices of intermediate layers to keep.
            prune_norm: Whether to prune normalization layer.
            prune_head: Whether to prune the classifier head.

        Returns:
            List of indices that were kept.
        Nr   r   )r   rÒ   rÛ   r	  )r*   r  r  r  r  r  s         r,   Úprune_intermediate_layersz#LowFormer.prune_intermediate_layers�  sZ   € õ  #7µs¸4¼;Ñ7GÔ7GÈÑ"QÔ"QÑˆ�iØ”k . 9¨q¡= .Ô1ˆŒØð 	%Ø×!Ò! !Ñ$Ô$Ð$ØÐr-   c                 óÐ   — |                       |¦  «        }| j        r4t          j                             ¦   «         st          | j        |¦  «        }n|                      |¦  «        }|S r1   )r×   rÑ   r9   r£   r  r   rÛ   r2   s     r,   Úforward_featureszLowFormer.forward_features³  sX   € Ø�IŠI�a‰LŒLˆØÔ"ð 	­5¬9×+AÒ+AÑ+CÔ+Cð 	Ý˜tœ{¨AÑ.Ô.ˆAˆAà—’˜A‘”ˆAØˆr-   Ú
pre_logitsc                 ó  — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }| j                             ¦   «         r|                     dddd¦  «        }|                      |¦  «        }| j        r!t          j        || j        | j	        ¬¦  «        n|}|s|  
                    |¦  «        }| j                             ¦   «         r*|                     dddd¦  «                             ¦   «         }|S )Nr   rL   r?   r   )r  )rà   rÁ   rß   Úis_identityr�   râ   rÃ   r‘   Údropoutr  rã   Ú
contiguous)r*   r.   r  s      r,   Úforward_headzLowFormer.forward_head»  só   € Ø�LŠL˜‰OŒOˆØ×Ò˜QÑÔˆØ�LŠL˜‰OŒOˆØÔ×'Ò'Ñ)Ô)ð 	&à—	’	˜!˜Q  1Ñ%Ô%ˆAØ×Ò Ñ"Ô"ˆØDHÄNÐY�AŒI�a˜œ°$´-Ð@Ñ@Ô@Ð@ÐXYˆØð 	#Ø—’ Ñ"Ô"ˆAØÔ×'Ò'Ñ)Ô)ð 	3Ø—	’	˜!˜Q  1Ñ%Ô%×0Ò0Ñ2Ô2ˆAØˆr-   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r1   )r  r"  r2   s     r,   r3   zLowFormer.forwardÊ  s-   € Ø×!Ò! !Ñ$Ô$ˆØ×Ò˜aÑ Ô ˆØˆr-   ©F)Tr1   )NFFr
  F)r   FT)$r4   r5   r6   r(   rV   r¹   r   rX   Ústrr   r8   r	   r7   rY   r   r'   rå   r9   r£   Úignorer   rû   r   r   rÿ   r  r  r	  r:   r
   r  r  r  r"  r3   r;   r<   s   @r,   r   r   ˆ  s   ø€ € € € € ð
 Ø#Ø$Ø+7Ø"Ø$'Ø$'Ø"#Ø*,¬.Ø)+¬Ø#ØØ!Ø"#Ø'(ØDHØ!"ØØð-}'ð }'à˜Sœ	ð}'ð ˜Sœ	ð}'ð ð	}'ð
 ð}'ð ð}'ð ˜s C˜xœð}'ð ð}'ð "ð}'ð "ð}'ð  ð}'ð ˜RœYœð}'ð ˜BœI”ð}'ð ð}'ð ð}'ð  ð!}'ð"  ð#}'ð$  %ð%}'ð& '/¨u°U¸C°ZÔ/@Ô&Að'}'ð( ð)}'ð }'ð }'ð }'ð }'ð }'ð~2 B¤Ið 2°$ð 2ð 2ð 2ð 2ð „YÔð ð ð ð ñ Ôðð „YÔðð  Dð °T¸#¸s¸(´^ð ð ð ñ Ôðð „YÔð)ð )¨Tð )ð )ð )ñ Ôð)ð „YÔð ¤	ð ð ð ñ Ôðð,ð ,¨Cð ,¸hÀs¼mð ,ð ,ð ,ð ,ð  8<ØØ$Ø$Ø',ð* ð * àŒ|ð* ð ˜e C¨¨c¬ NÔ3Ô4ð* ð ð	* ð
 ð* ð ð* ð !%ð* ð 
ˆt�E”LÔ! 5¨¬°t¸E¼LÔ7IÐ)IÔ#JÐJÔ	Kð* ð * ð * ð * ð\ ./Ø$Ø#ð	ð à˜3  S¤	˜>Ô*ðð ðð ð	ð
 
ˆcŒðð ð ð ð, %¤,ð °5´<ð ð ð ð ðð ˜eœlð ¸ð ÈÌð ð ð ð ð˜œð ¨%¬,ð ð ð ð ð ð ð ð r-   Ú
state_dictÚmodelr/   c                 ó  — d| v r| S |                       d| ¦  «        } i }|                      ¦   «         D �]U\  }}|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     d	d
¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|                     dd¦  «        }|||<   �ŒW|S ) Nzstem.0.conv.weightr'  zbackbone.input_stem.zstem.zbackbone.stages.rÊ   zhead.opseq.zhead.zhead.op_list.0.zin_conv.zhead.op_list.2.linear.zpre_classifier.0.zhead.op_list.2.norm.zpre_classifier.1.zhead.op_list.3.linear.zclassifier.z	.op_list.ú.z.total.0.0.z.attn.z.total.0.1.z.mlp.z	.total.0.z	.total.1.z.local.z.conv_proj.0.z.conv_proj.conv.z.conv_proj.1.z.conv_proj.norm.z	.pwise.0.z.pwise.)ÚgetÚitemsÚreplace)r'  r(  Úout_dictrž   rŸ   s        r,   Úcheckpoint_filter_fnr/  Ð  sƒ  € Ø˜zÐ)Ð)ØÐØ—’ ¨jÑ9Ô9€JØ€HØ× Ò Ñ"Ô"ð ñ ‰ˆˆ1à�IŠIÐ,¨gÑ6Ô6ˆØ�IŠIÐ(¨)Ñ4Ô4ˆð �IŠI�m WÑ-Ô-ˆØ�IŠIÐ'¨Ñ4Ô4ˆØ�IŠIÐ.Ð0CÑDÔDˆØ�IŠIÐ,Ð.AÑBÔBˆØ�IŠIÐ.°Ñ>Ô>ˆà�IŠI�k 3Ñ'Ô'ˆà�IŠI�m XÑ.Ô.ˆØ�IŠI�m WÑ-Ô-ˆØ�IŠI�k 8Ñ,Ô,ˆØ�IŠI�k 9Ñ-Ô-ˆà�IŠI�oÐ'9Ñ:Ô:ˆØ�IŠI�oÐ'9Ñ:Ô:ˆà�IŠI�k 9Ñ-Ô-ˆØˆ�‰‰Ø€Or-   Ú ÚurlÚkwargsc                 ó8   — | dddddt           t          dddd	d
œ|¥S )Nr»   )r?   éà   r4  )rt   rt   gffffffî?Úbicubiczstem.0.convrã   z)https://github.com/altair199797/LowFormerz
apache-2.0)r1  rÀ   Ú
input_sizeÚ	pool_sizeÚcrop_pctÚinterpolationÚmeanró   Ú
first_convrã   Ú
origin_urlÚlicenser   )r1  r2  s     r,   Ú_cfgr>  ð  s:   € à 4°}ÐSYØ¨9Ý%Õ.BØ#°<ØAÈlðð ð ðð r-   ztimm/)Ú	hf_hub_id)zlowformer_b0.in1kzlowformer_b1.in1kzlowformer_b15.in1kzlowformer_b2.untrainedzlowformer_b3.in1kzlowformer_e1.in1kzlowformer_e2.in1kzlowformer_e3.in1kFÚvariantÚ
pretrainedc                 ó\   — t          t          | |ft          t          dd¬¦  «        dœ|¤Ž}|S )N)r   r   rL   r?   T)Úout_indicesÚflatten_sequential)Úpretrained_filter_fnÚfeature_cfg)r   r   r/  rÚ   )r@  rA  r2  r(  s       r,   Ú_create_lowformerrG    sF   € Ý Ý�7˜Jðå1Ý \ÀdÐKÑKÔKðð ð ð	ð €Eð €Lr-   c           	      ó^   — t          g d¢g d¢¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )z'Instantiate LowFormer-B0 model variant.©é   é    é@   é€   r®   )r   r   r   r?   rs   )r½   r¾   Úlowformer_b0rA  )rN  ©rÚ   rG  ©rA  r2  Ú
model_argss      r,   rN  rN    sS   € õ Ø)Ð)Ð)°o°o°oðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   c           	      ó`   — t          g d¢g d¢d¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )z'Instantiate LowFormer-B1 model variant.rI  )r   r   r   r­   r­   ©r\   r\   r\   r\   )r½   r¾   rÇ   Úlowformer_b1rA  )rT  rO  rP  s      r,   rT  rT  (  sV   € õ Ø)Ð)Ð)°o°o°oØ!-ðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   c           	      ób   — t          g d¢g d¢dd¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )	z)Instantiate LowFormer-B1.5 model variant.©é   é(   éP   é    i@  ©r   r   r   r\   r\   ©i 	  i 
  rS  ©r½   r¾   rÂ   rÇ   Úlowformer_b15rA  )r^  rO  rP  s      r,   r^  r^  2  sY   € õ Ø)Ð)Ð)°o°o°oÐS_Ø!-ðñ ô €Jõ ÐbÐb¸ÐbÅtÈJÐGaÐGaÐZ`ÐGaÐGaÐbÐbÐbr-   c           	      ób   — t          g d¢g d¢dd¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )	z'Instantiate LowFormer-B2 model variant.)é   é0   é`   éÀ   i€  r[  r\  rS  r]  Úlowformer_b2rA  )rd  rO  rP  s      r,   rd  rd  <  sY   € õ Ø)Ð)Ð)°o°o°oÐS_Ø!-ðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   c           	      ób   — t          g d¢g d¢dd¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )	z'Instantiate LowFormer-B3 model variant.©rK  rL  rM  r®   i   ©r   rL   r?   r\   r\   rs   rS  )r½   r¾   rÆ   rÇ   Úlowformer_b3rA  )rh  rO  rP  s      r,   rh  rh  F  sX   € õ Ø*Ð*Ð*°°°Ø°lðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   c           	      óf   — t          g d¢g d¢dddd¬¦  «        }t          d	d| it          |fi |¤Ž¤ŽS )
z0Instantiate LowFormer-E1 edge GPU model variant.rV  )r   r   r   rs   rs   r\  FrS  )r½   r¾   rÂ   r¡   r©   rÇ   Úlowformer_e1rA  )rj  rO  rP  s      r,   rj  rj  P  s]   € õ Ø)Ð)Ð)°o°o°oÐS_Ø˜U¸\ðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   c           	      óf   — t          g d¢g d¢dddd¬¦  «        }t          d	d| it          |fi |¤Ž¤ŽS )
z0Instantiate LowFormer-E2 edge GPU model variant.rf  )r   rL   r?   rs   rs   Frs   rS  )r½   r¾   r¡   r©   rÆ   rÇ   Úlowformer_e2rA  )rl  rO  rP  s      r,   rl  rl  Z  s]   € õ Ø*Ð*Ð*°°°Ø˜U°aÐR^ðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   c           	      ód   — t          g d¢g d¢ddd¬¦  «        }t          d	d| it          |fi |¤Ž¤ŽS )
z0Instantiate LowFormer-E3 edge GPU model variant.rf  rg  Frs   rS  )r½   r¾   r©   rÆ   rÇ   Úlowformer_e3rA  )rn  rO  rP  s      r,   rn  rn  d  sZ   € õ Ø*Ð*Ð*°°°Ø¨!Àlðñ ô €Jõ ÐaÐa¸
ÐaÅdÈ:ÐF`ÐF`ÐY_ÐF`ÐF`ÐaÐaÐar-   )r0  r$  )?rU   Útypingr   r   r   r   r   r   r	   r
   r9   Útorch.nnr(   Útorch.nn.functionalÚ
functionalr‘   Ú	timm.datar   r   Útimm.layersr   r   r   r   r   r   r   r   Ú_builderr   Ú	_featuresr   Ú_features_fxr   Ú_manipulater   Ú	_registryr   r   Ú__all__r7   r   r>   r[   rk   rp   r§   r   r%  r:   r/  r>  Údefault_cfgsrY   rG  rN  rT  r^  rd  rh  rj  rl  rn  r¯   r-   r,   ú<module>r|     sl  ððð ð< FÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EÐ Eà €€€Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ Að	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð 	ð +Ð *Ð *Ð *Ð *Ð *Ø +Ð +Ð +Ð +Ð +Ð +Ø 1Ð 1Ð 1Ð 1Ð 1Ð 1Ø 'Ð 'Ð 'Ð 'Ð 'Ð 'Ø <Ð <Ð <Ð <Ð <Ð <Ð <Ð <àˆ-€ð?ð ?ð ?ð ?ð ?�B”Iñ ?ô ?ð ?ð$)ð )ð )ð )ð )�”	ñ )ô )ð )ðX<ð <ð <ð <ð <ˆRŒYñ <ô <ð <ð~-ð -ð -ð -ð -�"”)ñ -ô -ð -ðd ðRð Rð Rð Rð R�B”Iñ Rô Rñ ÔðRðjGð Gð Gð Gð G�R”Yñ Gô Gð GðTEð Eð Eð Eð E�”	ñ Eô Eð EðP
 T¨#¨u¬|Ð*;Ô%<ð ÀRÄYð ÐSWÐX[Ð]bÔ]iÐXiÔSjð ð ð ð ð@ð ˆcð  #ð ¨$¨s°C¨x¬.ð ð ð ð ð %Ð$Ø˜Øðñ ô ð ˜Øðñ ô ð ˜$Øðñ ô ð #˜d™fœfØ˜Øðñ ô ð ˜Øðñ ô ð ˜Øðñ ô ð ˜Øðñ ô ð)&ð &ñ ô €ð4ð ˜sð °ð Èð ÐPYð ð ð ð ð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð ðcð c˜dð c°cð c¸ið cð cð cñ „ðcð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð ðbð b˜Tð b°Sð b¸Yð bð bð bñ „ðbð bð br-   