§
    ·¯¸jõ²  ã                   ó¶  — d Z ddlmZ ddlmZmZ ddlmZ ddlm	Z	m
Z
mZmZmZ ddlZddl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mZmZmZ d	d
l m!Z! d	dl"m#Z# d	dl$m%Z% d	dl&m'Z'm(Z( ddgZ)e G d„ d¦  «        ¦   «         Z* G d„ dej+        ¦  «        Z,�dde-de.de
fd„Z/ G d„ dej+        ¦  «        Z0e# G d„ dej+        ¦  «        ¦   «         Z1	 	 	 	 	 	 �dde2de2d e-d!ee
         d"ee
         d#e3deej4        e2ee-e	f         f         fd$„Z5 e6dd%d&d'd(d)d*d+d,d-d.d/d0d1d2¬3¦  «        Z7 G d4„ dej+        ¦  «        Z8	 	 	 	 	 �dd7ee2d8f         d9ee2d8f         d:ee2         d"e-d;ee-         d<eee-e	f                  de*fd=„Z9�dd7ee2d8f         d9ee2d8f         de*fd?„Z:	 	 	 	 	 	 	 �dd7ee2d8f         d9ee2d8f         d:e2dFe.dGe.d"e-d;e-d<eee-e	f                  de*fdH„Z;	 	 	 �dd7ee2d8f         d9ee2d8f         d"e-dIe3de*f
dJ„Z< e6�di dK e<dL¬M¦  «        “dN e<dO¬M¦  «        “dP e<dQ¬M¦  «        “dR e<dS¬M¦  «        “dT e<dU¬M¦  «        “dV e<dW¬M¦  «        “dX e<dY¬M¦  «        “dZ e;dL¬M¦  «        “d[ e;dO¬M¦  «        “d\ e;dQ¬M¦  «        “d] e;dS¬M¦  «        “d^ e;dU¬M¦  «        “d_ e;dW¬M¦  «        “d` e;dY¬M¦  «        “da e;db¬M¦  «        “dc e;dLdddedf e6dfdg¬h¦  «        di¬j¦  «        “dk e;dLdddedfdl e6¦   «         di¬m¦  «        “dn e;dOdodedfdl e6¦   «         di¬m¦  «        “dp e;dQdodedfdl e6¦   «         di¬m¦  «        “dq e;dSdodedfdl e6¦   «         di¬m¦  «        “dr e:ds¬M¦  «        “dt e:du¬M¦  «        “dv e:dwdx¬y¦  «        “dz e:d{d|¬y¦  «        “d} e:d~d¬y¦  «        “d€ e:d�d‚¬y¦  «        “dƒ e9d„¬M¦  «        “d… e9d†¬M¦  «        “d‡ e9dˆ¬M¦  «        “d‰ e9d„dE e6dŠ¬‹¦  «        ¬Œ¦  «        “d� e9d†dE e6dŠ¬‹¦  «        ¬Œ¦  «        “dŽ e9dˆdE e6dŠ¬‹¦  «        ¬Œ¦  «        “d� e9d„dl e6¦   «         ¬Œ¦  «        “d� e9d†dl e6¦   «         ¬Œ¦  «        “d‘ e9dˆdl e6¦   «         ¬Œ¦  «        “d’ e;d“d”dddgdf e6dfdg¬h¦  «        di¬•¦  «        “ŽZ=�dd—e-d˜e3d™e	de8fdš„Z>�dd›e-d™e	dee-e	f         fdœ„Z? e'i d� e?dždŸd d¡d¢d£d¤¬¥¦  «        “d¦ e?džd§d¨d©dªd«d¤¬¥¦  «        “d¬ e?džd­d®d¢d¯d°d¤¬¥¦  «        “d± e?džd²d³dªd´dµd¤¬¥¦  «        “d¶ e?džd·d¸d¹dºd»d¤¬¥¦  «        “d¼ e?džd½d¾d´d¿dÀd¤¬¥¦  «        “dÁ e?dždÂdÃdÄdÅdÆd¤¬¥¦  «        “dZ e?dd d¡d¢¬Ç¦  «        “d[ e?dd¨d©dª¬Ç¦  «        “d\ e?dd®d¢d¯¬Ç¦  «        “d] e?dd³dªd´¬Ç¦  «        “d^ e?dd¸d¹dº¬Ç¦  «        “d_ e?dd¾d´d¿¬Ç¦  «        “d` e?ddÃdÄdÅ¬Ç¦  «        “da e?ddÈdÉdÊ¬Ç¦  «        “dË e?dždÌd¨d©dÍd¬Î¦  «        “dÏ e?dždÐd¨d©dÍd¬Î¦  «        “i dÑ e?dždÒd®d¢dªd¬Î¦  «        “dÓ e?dždÔd³dªd¹d¬Î¦  «        “dq e?ddÕd¯dÄd¬Ö¦  «        “dr e?dd d¡d¢d×¬Ø¦  «        “dÙ e?dždÚd®d¢dÍd×¬Û¦  «        “dv e?dd®dÜdÝd×¬Ø¦  «        “dz e?ddÞdÍdªd×¬Ø¦  «        “d} e?dd³dªd¹d×¬Ø¦  «        “d€ e?dd¸d¹dßd×¬Ø¦  «        “dƒ e?dd×¬à¦  «        “dá e?dždâd®d¢dÍdµd×¬ã¦  «        “d‡ e?dd×¬à¦  «        “d‰ e?dd×¬à¦  «        “d� e?dd×¬à¦  «        “dŽ e?dd×¬à¦  «        “d� e?dd×¬à¦  «        “d� e?dd×¬à¦  «        “¥ e?dd×¬à¦  «         e?dždädädådædç¬è¦  «        déœ¥¦  «        Z@e(�dd˜e3d™e	de8fdê„¦   «         ZAe(�dd˜e3d™e	de8fdë„¦   «         ZBe(�dd˜e3d™e	de8fdì„¦   «         ZCe(�dd˜e3d™e	de8fdí„¦   «         ZDe(�dd˜e3d™e	de8fdî„¦   «         ZEe(�dd˜e3d™e	de8fdï„¦   «         ZFe(�dd˜e3d™e	de8fdð„¦   «         ZGe(�dd˜e3d™e	de8fdñ„¦   «         ZHe(�dd˜e3d™e	de8fdò„¦   «         ZIe(�dd˜e3d™e	de8fdó„¦   «         ZJe(�dd˜e3d™e	de8fdô„¦   «         ZKe(�dd˜e3d™e	de8fdõ„¦   «         ZLe(�dd˜e3d™e	de8fdö„¦   «         ZMe(�dd˜e3d™e	de8fd÷„¦   «         ZNe(�dd˜e3d™e	de8fdø„¦   «         ZOe(�dd˜e3d™e	de8fdù„¦   «         ZPe(�dd˜e3d™e	de8fdú„¦   «         ZQe(�dd˜e3d™e	de8fdû„¦   «         ZRe(�dd˜e3d™e	de8fdü„¦   «         ZSe(�dd˜e3d™e	de8fdý„¦   «         ZTe(�dd˜e3d™e	de8fdþ„¦   «         ZUe(�dd˜e3d™e	de8fdÿ„¦   «         ZVe(�dd˜e3d™e	de8f�d „¦   «         ZWe(�dd˜e3d™e	de8f�d„¦   «         ZXe(�dd˜e3d™e	de8f�d„¦   «         ZYe(�dd˜e3d™e	de8f�d„¦   «         ZZe(�dd˜e3d™e	de8f�d„¦   «         Z[e(�dd˜e3d™e	de8f�d„¦   «         Z\e(�dd˜e3d™e	de8f�d„¦   «         Z]e(�dd˜e3d™e	de8f�d„¦   «         Z^e(�dd˜e3d™e	de8f�d„¦   «         Z_e(�dd˜e3d™e	de8f�d	„¦   «         Z`e(�dd˜e3d™e	de8f�d
„¦   «         Zae(�dd˜e3d™e	de8f�d„¦   «         Zbe(�dd˜e3d™e	de8f�d„¦   «         Zce(�dd˜e3d™e	de8f�d„¦   «         ZddS (  aæ   Normalization Free Nets. NFNet, NF-RegNet, NF-ResNet (pre-activation) Models

Paper: `Characterizing signal propagation to close the performance gap in unnormalized ResNets`
    - https://arxiv.org/abs/2101.08692

Paper: `High-Performance Large-Scale Image Recognition Without Normalization`
    - https://arxiv.org/abs/2102.06171

Official Deepmind JAX code: https://github.com/deepmind/deepmind-research/tree/master/nfnets

Status:
* These models are a work in progress, experiments ongoing.
* Pretrained weights for two models so far, more to come.
* Model details updated to closer match official JAX code now that it's released
* NF-ResNet, NF-RegNet-B, and NFNet-F models supported

Hacked together by / copyright Ross Wightman, 2021.
é    )ÚOrderedDict)Ú	dataclassÚreplace)Úpartial)ÚAnyÚCallableÚDictÚOptionalÚTupleN©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚClassifierHeadÚDropPathÚcalculate_drop_path_ratesÚAvgPool2dSameÚScaledStdConv2dÚScaledStdConv2dSameÚget_act_layerÚ
get_act_fnÚget_attnÚmake_divisibleÚget_device_dtypeé   )Úbuild_model_with_cfg)Úregister_notrace_module)Úcheckpoint_seq)Úgenerate_default_cfgsÚregister_modelÚNormFreeNetÚNfCfgc                   óš  — e Zd ZU dZeeeeef         ed<   eeeeef         ed<   dZeed<   dZ	e
ed<   dZee         ed	<   dZee         ed
<   dZee
         ed<   dZeee
ef                  ed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   dZeed<   d Ze
ed!<   dS )"r!   z.Configuration for Normalization-Free Networks.ÚdepthsÚchannelsgš™™™™™É?ÚalphaÚ3x3Ú	stem_typeNÚstem_chsÚ
group_sizeÚ
attn_layerÚattn_kwargsç       @Ú	attn_gainç      ð?Úwidth_factorç      à?Úbottle_ratior   Únum_featuresé   Úch_divFÚregÚ
extra_convÚgamma_in_actÚsame_paddinggñhãˆµøä>Ústd_conv_epsÚskipinitÚzero_init_fcÚsiluÚ	act_layer) Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   ÚintÚ__annotations__r%   Úfloatr'   Ústrr(   r
   r)   r*   r+   r	   r   r-   r/   r1   r2   r4   r5   Úboolr6   r7   r8   r9   r:   r;   r=   © ó    úa/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/nfnet.pyr!   r!   &   s’  € € € € € € à8Ð8Ø�#�s˜C Ð$Ô%Ð%Ð%Ñ%Ø�C˜˜c 3Ð&Ô'Ð'Ð'Ñ'Ø€Eˆ5ÐÐÑØ€IˆsÐÐÑØ"€Hˆh�sŒmÐ"Ð"Ñ"Ø $€J�˜”Ð$Ð$Ñ$Ø $€J�˜”Ð$Ð$Ñ$Ø,0€K�˜$˜s C˜xœ.Ô)Ð0Ð0Ñ0Ø€IˆuÐÐÑØ€L�%ÐÐÑØ€L�%ÐÐÑØ€L�#ÐÐÑØ€FˆC€O€O�OØ€CˆÐÐÑØ€J�ÐÐÑØ€L�$ÐÐÑØ€L�$ÐÐÑØ€L�%ÐÐÑØ€HˆdÐÐÑØ€L�$ÐÐÑØ€IˆsÐÐÑÐÐrH   c                   óV   ‡ — e Zd ZdZd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GammaActz.Activation function with gamma scaling factor.Úrelur.   FÚact_typeÚgammaÚinplacec                 óŒ   •— t          ¦   «                              ¦   «          t          |¦  «        | _        || _        || _        dS )zÕInitialize GammaAct.

        Args:
            act_type: Type of activation function.
            gamma: Scaling factor for activation output.
            inplace: Whether to perform activation in-place.
        N)ÚsuperÚ__init__r   Úact_fnrN   rO   )ÚselfrM   rN   rO   Ú	__class__s       €rI   rR   zGammaAct.__init__C   s<   ø€ õ 	‰Œ×ÒÑÔÐÝ  Ñ*Ô*ˆŒØˆŒ
ØˆŒˆˆrH   ÚxÚreturnc                 ój   — |                       || j        ¬¦  «                             | j        ¦  «        S )zzForward pass.

        Args:
            x: Input tensor.

        Returns:
            Scaled activation output.
        ©rO   )rS   rO   Úmul_rN   ©rT   rV   s     rI   ÚforwardzGammaAct.forwardP   s,   € ð �{Š{˜1 d¤lˆ{Ñ3Ô3×8Ò8¸¼ÑDÔDÐDrH   )rL   r.   F)r>   r?   r@   rA   rE   rD   rF   rR   ÚtorchÚTensorr\   Ú__classcell__©rU   s   @rI   rK   rK   @   s�   ø€ € € € € Ø8Ð8ðð  ð °eð ÈDð ð ð ð ð ð ð	E˜œð 	E¨%¬,ð 	Eð 	Eð 	Eð 	Eð 	Eð 	Eð 	Eð 	ErH   rK   r.   rM   rN   rW   c                 ó6   ‡ ‡— ddt           dt          fˆ ˆfd„}|S )zÝCreate activation function factory with gamma scaling.

    Args:
        act_type: Type of activation function.
        gamma: Scaling factor for activation output.

    Returns:
        Activation function factory.
    FrO   rW   c                 ó(   •— t          ‰‰| ¬¦  «        S )N)rN   rO   )rK   )rO   rM   rN   s    €€rI   Ú_createzact_with_gamma.<locals>._createf   s   ø€ Ý˜¨°wÐ?Ñ?Ô?Ð?rH   ©F)rF   rK   )rM   rN   rc   s   `` rI   Úact_with_gammare   \   sI   øø€ ð@ð @�ð @­(ð @ð @ð @ð @ð @ð @ð @à€NrH   c                   óz   ‡ — e Zd ZdZdddeddfdededededee         d	efˆ fd
„Zde	j
        de	j
        fd„Zˆ xZS )ÚDownsampleAvgzEAvgPool downsampling as in 'D' ResNet variants with dilation support.r   NÚin_chsÚout_chsÚstrideÚdilationÚfirst_dilationÚ
conv_layerc	                 ó8  •— t          ¦   «                              ¦   «          |dk    r|nd}	|dk    s|dk    r4|	dk    r|dk    rt          nt          j        }
 |
d|	dd¬¦  «        | _        nt          j        ¦   «         | _         |||dd||¬¦  «        | _        dS )a7  Initialize DownsampleAvg.

        Args:
            in_chs: Input channels.
            out_chs: Output channels.
            stride: Stride for downsampling.
            dilation: Dilation rate.
            first_dilation: First dilation rate (unused).
            conv_layer: Convolution layer type.
        r   é   TF)Ú	ceil_modeÚcount_include_pad)rj   ÚdeviceÚdtypeN)rQ   rR   r   ÚnnÚ	AvgPool2dÚpoolÚIdentityÚconv)rT   rh   ri   rj   rk   rl   rm   rr   rs   Ú
avg_strideÚavg_pool_fnrU   s              €rI   rR   zDownsampleAvg.__init__n   s§   ø€ õ* 	‰Œ×ÒÑÔÐØ'¨1š}˜}�V�V°!ˆ
Ø�AŠ:ˆ:˜ Aš˜Ø+5¸ª?¨?¸xÈ!º|¸|�-˜-ÕQSÔQ]ˆKØ#˜ A z¸TÐUZÐ[Ñ[Ô[ˆDŒIˆIåœ™œˆDŒIØ�J˜v w°¸!ÀFÐRWÐXÑXÔXˆŒ	ˆ	ˆ	rH   rV   rW   c                 óR   — |                       |                      |¦  «        ¦  «        S )ztForward pass.

        Args:
            x: Input tensor.

        Returns:
            Downsampled tensor.
        )rx   rv   r[   s     rI   r\   zDownsampleAvg.forwardŒ   s    € ð �yŠy˜Ÿš 1™œÑ&Ô&Ð&rH   )r>   r?   r@   rA   r   rB   r
   r   rR   r]   r^   r\   r_   r`   s   @rI   rg   rg   k   sÍ   ø€ € € € € ØOÐOð ØØ,0Ø#2ØØðYð YàðYð ðYð ð	Yð
 ðYð % SœMðYð !ðYð Yð Yð Yð Yð Yð<	'˜œð 	'¨%¬,ð 	'ð 	'ð 	'ð 	'ð 	'ð 	'ð 	'ð 	'rH   rg   c            %       óô   ‡ — e Zd ZdZddddddddddddddded	ddfd
edee         dededee         dedededee         dededededee	         dedee	         de	def$ˆ fd„Z
dej        dej        fd„Zˆ xZS ) ÚNormFreeBlockz-Normalization-Free pre-activation block.
    Nr   r.   ç      Ð?TFr,   ç        rh   ri   rj   rk   rl   r%   Úbetar1   r)   r4   r5   r6   r:   r*   r-   r=   rm   Údrop_path_ratec                 ó®  •— ||dœ}t          ¦   «                              ¦   «          |p|}|p|}t          |r||z  n||z  |
¦  «        }|	sdn||	z  }|	r|	|
z  dk    r|	|z  }|| _        || _        || _        ||k    s|dk    s||k    rt          ||f||||dœ|¤Ž| _        nd| _         |¦   «         | _         |||dfi |¤Ž| _	         |d¬¦  «        | _
         |||df|||d	œ|¤Ž| _        |r& |d¬¦  «        | _         |||dfd||d	œ|¤Ž| _        nd| _        d| _        |r|� ||fi |¤Ž| _        nd| _         |¦   «         | _         |||dfd
|rdndi|¤Ž| _        |s|� ||fi |¤Ž| _        nd| _        |dk    rt%          |¦  «        nt'          j        ¦   «         | _        |r#t'          j        t/          j        di |¤Ž¦  «        nd| _        dS )a„  Initialize NormFreeBlock.

        Args:
            in_chs: Input channels.
            out_chs: Output channels.
            stride: Stride for convolution.
            dilation: Dilation rate.
            first_dilation: First dilation rate.
            alpha: Alpha scaling factor for residual.
            beta: Beta scaling factor for pre-activation.
            bottle_ratio: Bottleneck ratio.
            group_size: Group convolution size.
            ch_div: Channel divisor for rounding.
            reg: Use RegNet-style configuration.
            extra_conv: Add extra 3x3 convolution.
            skipinit: Use skipinit initialization.
            attn_layer: Attention layer type.
            attn_gain: Attention gain factor.
            act_layer: Activation layer type.
            conv_layer: Convolution layer type.
            drop_path_rate: Stochastic depth drop rate.
        ©rr   rs   r   r   )rj   rk   rl   rm   NTrY   é   )rj   rk   ÚgroupsÚ	gain_initr.   r   )r   )rQ   rR   r   r%   r€   r-   rg   Ú
downsampleÚact1Úconv1Úact2Úconv2Úact2bÚconv2bÚattnÚact3Úconv3Ú	attn_lastr   rt   rw   Ú	drop_pathÚ	Parameterr]   ÚtensorÚskipinit_gain)rT   rh   ri   rj   rk   rl   r%   r€   r1   r)   r4   r5   r6   r:   r*   r-   r=   rm   r�   rr   rs   ÚddÚmid_chsr…   rU   s                           €rI   rR   zNormFreeBlock.__init__�   s´  ø€ ðZ ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ'Ð3¨8ˆØÐ#˜Vˆå ¸#Ð!Y ¨,Ñ!6Ð!6À7È\ÑCYÐ[aÑbÔbˆØ$Ð?��¨'°ZÑ*?ˆØð 	*˜* vÑ-°Ò2Ð2Ø  6Ñ)ˆGØˆŒ
ØˆŒ	Ø"ˆŒà�WÒÐ ¨!¢ ¨x¸>Ò/IÐ/IÝ+ØØðð Ø!Ø-Ø%ðð ð ðð ˆDŒOˆOð #ˆDŒOà�I‘K”KˆŒ	Ø�Z ¨°Ð9Ð9°bÐ9Ð9ˆŒ
Ø�I dÐ+Ñ+Ô+ˆŒ	Ø�Z ¨°!Ðq¸FÈ^ÐdjÐqÐqÐnpÐqÐqˆŒ
Øð 	Ø"˜¨4Ð0Ñ0Ô0ˆDŒJØ$˜* W¨g°qÐkÀÈXÐ^dÐkÐkÐhjÐkÐkˆDŒKˆKàˆDŒJØˆDŒKØð 	�:Ð)Ø"˜
 7Ð1Ð1¨bÐ1Ð1ˆDŒIˆIàˆDŒIØ�I‘K”KˆŒ	Ø�Z ¨°!Ð\Ð\ÀXÐ?U¸r¸rÐSUÐ\ÐY[Ð\Ð\ˆŒ
Øð 	"�zÐ-Ø'˜Z¨Ð6Ð6°2Ð6Ð6ˆDŒNˆNà!ˆDŒNØ5CÀaÒ5GÐ5G� .Ñ1Ô1Ð1ÍRÌ[É]Ì]ˆŒØEMÐW�Rœ\­%¬,Ð*@Ð*@¸RÐ*@Ð*@ÑAÔAÐAÐSWˆÔÐÐrH   rV   rW   c                 óº  — |                       |¦  «        | j        z  }|}| j        �|                      |¦  «        }|                      |¦  «        }|                      |                      |¦  «        ¦  «        }| j        �(|                      |                      |¦  «        ¦  «        }| j        �| j	        |                      |¦  «        z  }|  
                    |                      |¦  «        ¦  «        }| j        �| j	        |                      |¦  «        z  }|                      |¦  «        }| j        �|                     | j        ¦  «         || j        z  |z   }|S )zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        )rˆ   r€   r‡   r‰   r‹   rŠ   r�   rŒ   rŽ   r-   r�   r�   r‘   r’   r•   rZ   r%   )rT   rV   ÚoutÚshortcuts       rI   r\   zNormFreeBlock.forwardû   s'  € ð �iŠi˜‰lŒl˜TœYÑ&ˆð ˆØŒ?Ð&Ø—’ sÑ+Ô+ˆHð �jŠj˜‰oŒoˆØ�jŠj˜Ÿš 3™œÑ(Ô(ˆØŒ;Ð"Ø—+’+˜dŸjšj¨™oœoÑ.Ô.ˆCØŒ9Ð Ø”. 4§9¢9¨S¡>¤>Ñ1ˆCØ�jŠj˜Ÿš 3™œÑ(Ô(ˆØŒ>Ð%Ø”. 4§>¢>°#Ñ#6Ô#6Ñ6ˆCØ�nŠn˜SÑ!Ô!ˆàÔÐ)Ø�HŠH�TÔ'Ñ(Ô(Ð(Ø�D”JÑ Ñ)ˆØˆ
rH   )r>   r?   r@   rA   r   rB   r
   rD   rF   r   rR   r]   r^   r\   r_   r`   s   @rI   r}   r}   ˜   s¨  ø€ € € € € ðð ð &*ØØØ,0ØØØ"&Ø(,ØØØ$Ø"Ø-1Ø"Ø,0Ø#2Ø$&ØØð+\Xð \Xàð\Xð ˜c”]ð\Xð ð	\Xð
 ð\Xð % SœMð\Xð ð\Xð ð\Xð  ð\Xð ! œð\Xð ð\Xð ð\Xð ð\Xð ð\Xð ! Ô*ð\Xð  ð!\Xð"   Ô)ð#\Xð$ !ð%\Xð& "ð'\Xð \Xð \Xð \Xð \Xð \Xð|˜œð ¨%¬,ð ð ð ð ð ð ð ð rH   r}   Ú Trh   ri   r'   rm   r=   Úpreact_featurec                 óÄ  — ||dœ}d}	t          |dd¬¦  «        }
t          ¦   «         }|dv sJ ‚d|v rÔd|v r/d|vsJ ‚|d	z  |d
z  |dz  |f}d}d
}	t          |dz  dd¬¦  «        }
n5d|v rd|z  d	z  |dz  |f}n|dz  |dz  |f}d}t          |dz  dd¬¦  «        }
t          |¦  «        dz
  }t          t	          ||¦  «        ¦  «        D ];\  }\  }} || |fd|dœ|¤Ž|d|dz   › �<   ||k    r |d¬¦  «        |d|dz   › �<   |} Œ<n%d|v r || |fdddœ|¤Ž|d<   n || |fdddœ|¤Ž|d<   d|v rt          j        ddd¬¦  «        |d<   d
}	t          j        |¦  «        |	|
fS )a¬  Create stem module for NFNet models.

    Args:
        in_chs: Input channels.
        out_chs: Output channels.
        stem_type: Type of stem ('', 'deep', 'deep_tiered', 'deep_quad', '3x3', '7x7', etc.).
        conv_layer: Convolution layer type.
        act_layer: Activation layer type.
        preact_feature: Use pre-activation feature.

    Returns:
        Tuple of (stem_module, stem_stride, stem_feature_info).
    rƒ   ro   ú	stem.conv©Únum_chsÚ	reductionÚmodule)	r›   ÚdeepÚdeep_tieredÚ	deep_quadr&   Ú7x7Ú	deep_poolÚ3x3_poolÚ7x7_poolr£   Úquadrv   r3   é   )ro   r   r   ro   z
stem.conv3Útieredr„   )ro   r   r   z
stem.conv2r   )Úkernel_sizerj   rx   TrY   Úactr&   é   )rj   Úpadding)Údictr   ÚlenÚ	enumerateÚziprt   Ú	MaxPool2dÚ
Sequential)rh   ri   r'   rm   r=   rœ   rr   rs   r–   Ústem_strideÚstem_featureÚstemr(   ÚstridesÚlast_idxÚiÚcÚss                     rI   Úcreate_stemr¿     s:  € ð.  UÐ	+Ð	+€BØ€KÝ °1¸[ÐIÑIÔI€LÝ‰=Œ=€DØÐsÐsÐsÐsÐsØ�ÐÐØ�YÐÐà Ð*Ð*Ð*Ð*Ø 1™ g°¡l°G¸q±LÀ'ÐJˆHØ"ˆGØˆKÝ¨°1©ÀÈ,ÐWÑWÔWˆLˆLà˜9Ð$Ð$Ø ™K¨1Ñ,¨g¸©l¸GÐD��à# q™L¨'°Q©,¸Ð@�ØˆGÝ¨°1©ÀÈ,ÐWÑWÔWˆLÝ�x‘=”= 1Ñ$ˆÝ"¥3 x°Ñ#9Ô#9Ñ:Ô:ð 	ð 	‰IˆA‰v��1Ø#- :¨f°aÐ#WÀQÈqÐ#WÐ#WÐTVÐ#WÐ#WˆD�˜˜A™��Ñ Ø�HŠ}ˆ}Ø&/ i¸Ð&=Ñ&=Ô&=��]˜1˜q™5�]�]Ñ#ØˆFˆFð		ð
 
�)Ð	Ð	à!�z &¨'ÐQ¸qÈÐQÐQÈbÐQÐQˆˆV‰ˆð "�z &¨'ÐQ¸qÈÐQÐQÈbÐQÐQˆˆV‰à�ÐÐÝ”| A¨a¸Ð;Ñ;Ô;ˆˆV‰ØˆåŒ=˜ÑÔ ¨\Ð9Ð9rH   g   `·Uô?g   ÀyXô?g    \9û?g   `aKû?g   À¦µþ?g   €Ú ð?g    `lû?g   `iû?g   à|ð?g     7@g   €-œü?g    øµ@g   `®¹þ?g    ©€ù?)ÚidentityÚceluÚeluÚgeluÚ
leaky_reluÚlog_sigmoidÚlog_softmaxrL   Úrelu6ÚseluÚsigmoidr<   ÚsoftsignÚsoftplusÚtanhc                   ó   ‡ — e Zd ZdZ	 	 	 	 	 	 	 	 dded	ed
edededededefˆ fd„Z	e
j        j        d dedeeef         fd„¦   «         Ze
j        j        d!deddfd„¦   «         Ze
j        j        dej        fd„¦   «         Zd"d	edee         ddfd„Zde
j        de
j        fd„Zd de
j        dede
j        fd„Zde
j        de
j        fd„Zˆ xZS )#r    a*   Normalization-Free Network

    As described in :
    `Characterizing signal propagation to close the performance gap in unnormalized ResNets`
        - https://arxiv.org/abs/2101.08692
    and
    `High-Performance Large-Scale Image Recognition Without Normalization` - https://arxiv.org/abs/2102.06171

    This model aims to cover both the NFRegNet-Bx models as detailed in the paper's code snippets and
    the (preact) ResNet models described earlier in the paper.

    There are a few differences:
        * channels are rounded to be divisible by 8 by default (keep tensor core kernels happy),
            this changes channel dim and param counts slightly from the paper models
        * activation correcting gamma constants are moved into the ScaledStdConv as it has less performance
            impact in PyTorch when done with the weight scaling there. This likely wasn't a concern in the JAX impl.
        * a config option `gamma_in_act` can be enabled to not apply gamma in StdConv as described above, but
            apply it in each activation. This is slightly slower, numerically different, but matches official impl.
        * skipinit is disabled by default, it seems to have a rather drastic impact on GPU memory use and throughput
            for what it is/does. Approx 8-10% throughput loss.
    éè  r„   Úavgé    r   NÚcfgÚnum_classesÚin_chansÚglobal_poolÚoutput_strideÚ	drop_rater�   Úkwargsc
                 óX
  •— t          ¦   «                              ¦   «          ||	dœ}|| _        || _        || _        d| _        t          |fi |
¤Ž}|j        t          v sJ d|j        › d�¦   «         ‚|j	        rt          nt          }|j        r=t          |j        t          |j                 ¬¦  «        }t          ||j        ¬¦  «        }n;t!          |j        ¦  «        }t          |t          |j                 |j        ¬¦  «        }|j        r$t          t%          |j        ¦  «        fi |j        ¤Žnd}t)          |j        p|j        d	         |j        z  |j        ¦  «        }t3          |||j        f||d
œ|¤Ž\  | _        }}|g| _        t;          ||j        d¬¦  «        }|}|}d}d}g }t?          |j        ¦  «        D �]k\  }}|d	k    r|dk    rdnd}||k    r|dk    r||z  }d}||z  }|dv rdnd}g }tA          |j        |         ¦  «        D ]â}|d	k    o|d	k    }t)          |j        |         |j        z  |j        ¦  «        }|tC          d0i d|“d|“d|j"        “dd|dz  z  “d|d	k    r|nd“d|“d|“d|j#        “d|j$        r|rdn|j%        “d|j        “d|j$        “d|j&        “d|j'        “d|“d |j(        “d!|“d"|“d#||         |         “|¤Žgz  }|d	k    rd}||j"        dz  z  }|}|}Œã| xj        tS          ||d$|› �¬%¦  «        gz  c_        |tU          j+        |Ž gz  }�ŒmtU          j+        |Ž | _,        |j-        r\t)          |j        |j-        z  |j        ¦  «        | _-         ||| j-        dfi |¤Ž| _.        tS          | j-        |d&¬%¦  «        | j        d'<   n|| _-        tU          j/        ¦   «         | _.         ||j-        d	k    ¬(¦  «        | _0        | j-        | _1        te          | j-        |f|| j        d)œ|¤Ž| _3        |  4                    ¦   «         D �]\  } }!d*| v r˜tk          |!tT          j6        ¦  «        r~|j7        r%tT          j8         9                    |!j:        ¦  «         n&tT          j8         ;                    |!j:        d+d,¦  «         |!j<        �$tT          j8         9                    |!j<        ¦  «         Œ¢tk          |!tT          j=        ¦  «        rRtT          j8         >                    |!j:        d-d.¬/¦  «         |!j<        �$tT          j8         9                    |!j<        ¦  «         �ŒdS )1aÁ  
        Args:
            cfg: Model architecture configuration.
            num_classes: Number of classifier classes.
            in_chans: Number of input channels.
            global_pool: Global pooling type.
            output_stride: Output stride of network, one of (8, 16, 32).
            drop_rate: Dropout rate.
            drop_path_rate: Stochastic depth drop-path rate.
            **kwargs: Extra kwargs overlayed onto cfg.
        rƒ   Fz3Please add non-linearity constants for activation (z).)rN   )Úeps)rN   rÙ   Nr   )rm   r=   T)Ú	stagewiser   r.   ro   )r   ro   rh   ri   r%   r€   r0   rj   rk   rl   r)   r1   r4   r5   r6   r:   r*   r-   r=   rm   r�   zstages.rŸ   Ú
final_convéÿÿÿÿrY   )Ú	pool_typerÖ   Úfcr   g{®Gáz„?Úfan_inÚlinear)ÚmodeÚnonlinearityrG   )?rQ   rR   rÒ   rÓ   rÖ   Úgrad_checkpointingr   r=   Ú_nonlin_gammar8   r   r   r7   re   r   r9   r   r*   r   r+   r   r(   r$   r/   r4   r¿   r'   r¹   Úfeature_infor   r#   r³   Úranger}   r%   r)   r5   r1   r6   r:   r-   r±   rt   r¶   Ústagesr2   rÛ   rw   Ú	final_actÚhead_hidden_sizer   ÚheadÚnamed_modulesÚ
isinstanceÚLinearr;   ÚinitÚzeros_ÚweightÚnormal_ÚbiasÚConv2dÚkaiming_normal_)#rT   rÑ   rÒ   rÓ   rÔ   rÕ   rÖ   r�   rr   rs   r×   r–   rm   r=   r*   r(   r·   Ú	stem_featÚdrop_path_ratesÚprev_chsÚ
net_striderk   Úexpected_varrç   Ú	stage_idxÚstage_depthrj   rl   ÚblocksÚ	block_idxÚfirst_blockri   ÚnÚmrU   s#                                     €rI   rR   zNormFreeNet.__init__†  s4  ø€ õ0 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ&ˆÔØ ˆŒØ"ˆŒØ"'ˆÔå�cÐ$Ð$˜VÐ$Ð$ˆØŒ}¥Ð-Ð-Ð-Ð/vÐehÔerÐ/vÐ/vÐ/vÑ-Ô-Ð-Ø,/Ô,<ÐQÕ(Ð(Å/ˆ
ØÔð 	gÝ& s¤}½MÈ#Ì-Ô<XÐYÑYÔYˆIÝ  °Ô1AÐBÑBÔBˆJˆJå% c¤mÑ4Ô4ˆIÝ  µ=ÀÄÔ3OÐUXÔUeÐfÑfÔfˆJØMPÌ^Ðe•W�X c¤nÑ5Ô5ÐIÐI¸¼ÐIÐIÐIÐaeˆ
å! 3¤<Ð#B°3´<À´?ÀcÔFVÑ"VÐX[ÔXbÑcÔcˆÝ,7ØØØŒMð-
ð "Øð-
ð -
ð ð-
ð -
Ñ)ˆŒ	�; 	ð '˜KˆÔÝ3°NÀCÄJÐZ^Ð_Ñ_Ô_ˆØˆØ ˆ
ØˆØˆØˆÝ&/°´
Ñ&;Ô&;ð &	/ñ &	/Ñ"ˆI�{Ø# qš.˜.¨[¸1ª_¨_�Q�QÀ!ˆFØ˜]Ò*Ð*¨v¸ªz¨zØ˜FÑ"�Ø�Ø˜&Ñ ˆJØ"*¨fÐ"4Ð"4˜Q˜Q¸!ˆNàˆFÝ" 3¤:¨iÔ#8Ñ9Ô9ð #ð #�	Ø'¨1šnÐ?°¸a²�Ý(¨¬°iÔ)@À3ÔCSÑ)SÐUXÔU_Ñ`Ô`�Ø�=ð ð ð Ø#˜8ðØ-4¨Wðàœ)˜)ðð ˜l¨cÑ1Ñ1Ð1ðð &/°!¢^ ^˜6˜6¸øð	ð
 &˜Xðð $2 >ðð  #œ~˜~ðð (+¤wÐ!T°;Ð!T  ÀCÔDTøðð œ:˜:ðð œ˜ðð  #œ~˜~ðð !œ\˜\ðð  *˜zðð "œm˜mðð (˜iðð   *˜zð!ð" $3°9Ô#=¸iÔ#HÐ#HØð%ð ð ñ �ð(  ’>�>Ø#%�LØ ¤	¨Q¡Ñ.�Ø!)�Ø"��ØÐÔ¥$¨xÀ:ÐVkÐ`iÐVkÐVkÐ"lÑ"lÔ"lÐ!mÑmÐÔØ•r”} fÐ-Ð.Ñ.ˆF‰FÝ”m VÐ,ˆŒàÔð 	,å .¨sÔ/?À#ÔBRÑ/RÐTWÔT^Ñ _Ô _ˆDÔØ(˜j¨°4Ô3DÀaÐNÐNÈ2ÐNÐNˆDŒOÝ$(°Ô1BÈjÐanÐ$oÑ$oÔ$oˆDÔ˜bÑ!Ð!à (ˆDÔÝ œk™mœmˆDŒOØ"˜¨3Ô+;¸aÒ+?Ð@Ñ@Ô@ˆŒà $Ô 1ˆÔÝ"ØÔØð
ð "Ø”nð	
ð 
ð
 ð
ð 
ˆŒ	ð ×&Ò&Ñ(Ô(ð 	+ñ 	+‰DˆAˆqØ�qˆyˆy�Z¨­2¬9Ñ5Ô5ˆyØÔ#ð 7Ý”G—N’N 1¤8Ñ,Ô,Ð,Ð,å”G—O’O A¤H¨b°#Ñ6Ô6Ð6Ø”6Ð%Ý”G—N’N 1¤6Ñ*Ô*Ð*øÝ˜A�rœyÑ)Ô)ð +Ý”×'Ò'¨¬°xÈhÐ'ÑWÔWÐWØ”6Ð%Ý”G—N’N 1¤6Ñ*Ô*Ð*ùð	+ð 	+rH   FÚcoarserW   c                 ó8   — t          d|rdnddfdg¬¦  «        }|S )z"Group parameters for optimization.z^stemz^stages\.(\d+)z^stages\.(\d+)\.(\d+)N)z^final_conv)iŸ† )r¹   rü   )r±   )rT   r  Úmatchers      rI   Úgroup_matcherzNormFreeNet.group_matcher
  s>   € õ Øà&,ÐJÐ"Ð"Ð2JÈDÐQØ*ðð
ñ 
ô 
ˆð ˆrH   TÚenablec                 ó   — || _         dS )z)Enable or disable gradient checkpointing.N)rã   )rT   r  s     rI   Úset_grad_checkpointingz"NormFreeNet.set_grad_checkpointing  s   € ð #)ˆÔÐÐrH   c                 ó   — | j         j        S )zGet the classifier head.)rê   rÞ   )rT   s    rI   Úget_classifierzNormFreeNet.get_classifier  s   € ð ŒyŒ|ÐrH   c                 óZ   — t          | ¦  «        }|| _         | j        j        ||fi |¤Ž dS )zŸReset the classifier head.

        Args:
            num_classes: Number of classes for new classifier.
            global_pool: Global pooling type.
        N)r   rÒ   rê   Úreset)rT   rÒ   rÔ   r–   s       rI   Úreset_classifierzNormFreeNet.reset_classifier   s=   € õ ˜dÑ#Ô#ˆØ&ˆÔØˆŒ	Œ˜ [Ð7Ð7°BÐ7Ð7Ð7Ð7Ð7rH   rV   c                 ó$  — |                       |¦  «        }| j        r4t          j                             ¦   «         st          | j        |¦  «        }n|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|S )z’Forward pass through feature extraction layers.

        Args:
            x: Input tensor.

        Returns:
            Feature tensor.
        )	r¹   rã   r]   ÚjitÚis_scriptingr   rç   rÛ   rè   r[   s     rI   Úforward_featureszNormFreeNet.forward_features+  s|   € ð �IŠI�a‰LŒLˆØÔ"ð 	­5¬9×+AÒ+AÑ+CÔ+Cð 	Ý˜tœ{¨AÑ.Ô.ˆAˆAà—’˜A‘”ˆAØ�OŠO˜AÑÔˆØ�NŠN˜1ÑÔˆØˆrH   Ú
pre_logitsc                 ó^   — |r|                       ||¬¦  «        n|                       |¦  «        S )zàForward pass through classifier head.

        Args:
            x: Input features.
            pre_logits: Return features before final linear layer.

        Returns:
            Classification logits or features.
        )r  )rê   )rT   rV   r  s      rI   Úforward_headzNormFreeNet.forward_head=  s0   € ð 7AÐRˆt�yŠy˜ zˆyÑ2Ô2Ð2ÀdÇiÂiÐPQÁlÄlÐRrH   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S )zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output logits.
        )r  r  r[   s     rI   r\   zNormFreeNet.forwardI  s/   € ð ×!Ò! !Ñ$Ô$ˆØ×Ò˜aÑ Ô ˆØˆrH   )rÎ   r„   rÏ   rÐ   r   r   NNrd   )T)N)r>   r?   r@   rA   r!   rB   rE   rD   r   rR   r]   r  ÚignorerF   r	   r  r  rt   ÚModuler	  r
   r  r^   r  r  r\   r_   r`   s   @rI   r    r    p  s&  ø€ € € € € ðð ð0  $ØØ$Ø!#Ø!Ø$&ØØðB+ð B+àðB+ð ðB+ð ð	B+ð
 ðB+ð ðB+ð ðB+ð "ðB+ð ðB+ð B+ð B+ð B+ð B+ð B+ðH „YÔð	ð 	 Dð 	°T¸#¸s¸(´^ð 	ð 	ð 	ñ Ôð	ð „YÔð)ð )¨Tð )¸Tð )ð )ð )ñ Ôð)ð „YÔð ¤	ð ð ð ñ Ôðð	8ð 	8¨Cð 	8¸hÀs¼mð 	8ÐW[ð 	8ð 	8ð 	8ð 	8ð %¤,ð °5´<ð ð ð ð ð$
Sð 
S˜eœlð 
S¸ð 
SÈÌð 
Sð 
Sð 
Sð 
Sð˜œð ¨%¬,ð ð ð ð ð ð ð ð rH   ©é   é   i   i   rL   r#   .r$   r)   r*   r+   c                 ó>   — |pi }t          | |ddd||||¬¦	  «	        }|S )ar  Create NFNet ResNet configuration.

    Args:
        depths: Number of blocks in each stage.
        channels: Channel dimensions for each stage.
        group_size: Group convolution size.
        act_layer: Activation layer type.
        attn_layer: Attention layer type.
        attn_kwargs: Attention layer arguments.

    Returns:
        NFNet configuration.
    r©   é@   r~   )	r#   r$   r'   r(   r1   r)   r=   r*   r+   )r!   )r#   r$   r)   r=   r*   r+   rÑ   s          rI   Ú
_nfres_cfgr  W  sD   € ð* Ð# €KÝ
ØØØØØØØØØð
ñ 
ô 
€Cð €JrH   ©é0   éh   éÐ   é¸  c                 ót   — d|d         z  dz  }t          d¬¦  «        }t          | |dddd	|d
d|¬¦
  «
        }|S )zÂCreate NFNet RegNet configuration.

    Args:
        depths: Number of blocks in each stage.
        channels: Channel dimensions for each stage.

    Returns:
        NFNet configuration.
    i   rÜ   r!  r0   ©Úrd_ratior&   r3   g      è?g      @TÚse)
r#   r$   r'   r)   r/   r1   r2   r5   r*   r+   )r±   r!   )r#   r$   r2   r+   rÑ   s        rI   Ú
_nfreg_cfgr&  {  s`   € ð ˜( 2œ,Ñ&¨#Ñ-€LÝ Ð$Ñ$Ô$€KÝ
ØØØØØØØ!ØØØðñ ô €Cð €JrH   ©r  r  é   r(  é€   r0   r,   rÃ   r%  r1   Ú	feat_multc                 ó’   — t          |d         |z  ¦  «        }|�|nt          d¬¦  «        }t          | |dd||d||||¬¦  «        }	|	S )	aÊ  Create NFNet configuration.

    Args:
        depths: Number of blocks in each stage.
        channels: Channel dimensions for each stage.
        group_size: Group convolution size.
        bottle_ratio: Bottleneck ratio.
        feat_mult: Feature multiplier for final layer.
        act_layer: Activation layer type.
        attn_layer: Attention layer type.
        attn_kwargs: Attention layer arguments.

    Returns:
        NFNet configuration.
    rÜ   Nr0   r#  r¥   r)  T)r#   r$   r'   r(   r)   r1   r6   r2   r=   r*   r+   )rB   r±   r!   )
r#   r$   r)   r1   r*  r=   r*   r+   r2   rÑ   s
             rI   Ú
_nfnet_cfgr,  –  so   € õ2 �x ”| iÑ/Ñ0Ô0€LØ!,Ð!8�+�+½dÈCÐ>PÑ>PÔ>P€KÝ
ØØØØØØ!ØØ!ØØØðñ ô €Cð €JrH   r:   c                 óˆ   — t          | |ddddddd|t          |d         dz  ¦  «        |dt          d¬¦  «        ¬	¦  «        }|S )
a  Create DeepMind NFNet configuration.

    Args:
        depths: Number of blocks in each stage.
        channels: Channel dimensions for each stage.
        act_layer: Activation layer type.
        skipinit: Use skipinit initialization.

    Returns:
        NFNet configuration.
    r¥   r)  r0   TrÜ   r,   r%  r#  )r#   r$   r'   r(   r)   r1   r6   r7   r8   r:   r2   r=   r*   r+   )r!   rB   r±   )r#   r$   r=   r:   rÑ   s        rI   Ú_dm_nfnet_cfgr.  Á  se   € õ" ØØØØØØØØØØÝ˜ "œ¨Ñ+Ñ,Ô,ØØÝ #Ð&Ñ&Ô&ðñ ô €Cð  €JrH   Údm_nfnet_f0)r   ro   é   r„   )r#   Údm_nfnet_f1)ro   r«   é   r0  Údm_nfnet_f2)r„   r0  é   é	   Údm_nfnet_f3)r«   r3   é   r2  Údm_nfnet_f4)é   é
   é   é   Údm_nfnet_f5)r0  r2  é$   r4  Údm_nfnet_f6)r¯   é   é*   é   Únfnet_f0Únfnet_f1Únfnet_f2Únfnet_f3Únfnet_f4Únfnet_f5Únfnet_f6Únfnet_f7)r3   é   r  r7  Únfnet_l0g      ø?r  r~   r3   )r$  Ú
rd_divisorr<   )r#   r*  r)   r1   r+   r=   Úeca_nfnet_l0Úeca)r#   r*  r)   r1   r*   r+   r=   Úeca_nfnet_l1ro   Úeca_nfnet_l2Úeca_nfnet_l3Únf_regnet_b0)r   r„   r0  r0  Únf_regnet_b1)ro   r«   r¯   r¯   Únf_regnet_b2)ro   r«   r3   r3   )é8   ép   éè   iè  )r#   r$   Únf_regnet_b3)ro   r9  r5  r5  )rV  r)  éø   i  Únf_regnet_b4)ro   r0  é   r\  )r  é�   é   ih  Únf_regnet_b5)r„   r¯   r@  r@  )éP   é¨   iP  iÀ  Únf_resnet26)ro   ro   ro   ro   Únf_resnet50)r„   r«   r0  r„   Únf_resnet101)r„   r«   é   r„   Únf_seresnet26g      °?r#  )r#   r*   r+   Únf_seresnet50Únf_seresnet101Únf_ecaresnet26Únf_ecaresnet50Únf_ecaresnet101Ú
test_nfnet)r   r   r   r   )rÐ   r  é`   r)  )r#   r$   r*  r)   r1   r+   r=   FÚvariantÚ
pretrainedr×   c                 ój   — t           |          }t          d¬¦  «        }t          t          | |f||dœ|¤ŽS )zÝCreate a NormFreeNet model.

    Args:
        variant: Model variant name.
        pretrained: Load pretrained weights.
        **kwargs: Additional model arguments.

    Returns:
        NormFreeNet model instance.
    T)Úflatten_sequential)Ú	model_cfgÚfeature_cfg)Ú
model_cfgsr±   r   r    )rn  ro  r×   rr  rs  s        rI   Ú_create_normfreenetru  &  sU   € õ ˜7Ô#€IÝ¨$Ð/Ñ/Ô/€KÝÝØØðð Øðð ð ðð ð rH   Úurlc                 ó6   — | dddddt           t          dddd	œ|¥S )
z»Create default configuration dictionary.

    Args:
        url: Model weight URL.
        **kwargs: Additional configuration options.

    Returns:
        Configuration dictionary.
    rÎ   ©r„   éà   ry  ©r¯   r¯   çÍÌÌÌÌÌì?Úbicubicz
stem.conv1zhead.fcz
apache-2.0)rv  rÒ   Ú
input_sizeÚ	pool_sizeÚcrop_pctÚinterpolationÚmeanÚstdÚ
first_convÚ
classifierÚlicenser   )rv  r×   s     rI   Ú_dcfgr†  =  s9   € ð Ø¨=ÀvØ¨)Ý%Õ.BØ"°)Èðð ð ðð rH   zdm_nfnet_f0.dm_in1kztimm/zmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f0-604f9c3a.pth)r0  r0  )r„   éÀ   r‡  )r„   r  r  r{  Úsquash)Ú	hf_hub_idrv  r~  r}  Útest_input_sizer  Ú	crop_modezdm_nfnet_f1.dm_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f1-fc540f82.pthrz  rx  )r„   é@  rŒ  g…ëQ¸í?zdm_nfnet_f2.dm_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f2-89875923.pth)r3   r3   )r„   é`  r�  gq=
×£pí?zdm_nfnet_f3.dm_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f3-d74ab3aa.pth)r:  r:  )r„   é   rŽ  g®Gázî?zdm_nfnet_f4.dm_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f4-0ac5b10b.pth)r2  r2  )r„   é€  r�  )r„   r  r  g;ßO�—nî?zdm_nfnet_f5.dm_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f5-ecb20ab1.pth)é   r�  )r„   é   r‘  gºI+‡î?zdm_nfnet_f6.dm_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-dnf-weights/dm_nfnet_f6-e0f12116.pth)r@  r@  )r„   éÀ  r’  )r„   é@  r“  gd;ßO�—î?)rv  r~  r}  rŠ  )r<  r<  )r„   éà  r”  )r„   é`  r•  znfnet_l0.ra2_in1kzjhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/nfnet_l0_ra2-45c6688d.pth)r„   r^  r^  )r‰  rv  r~  r}  rŠ  Útest_crop_pctzeca_nfnet_l0.ra2_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ecanfnet_l0_ra2-e3e9ac50.pthzeca_nfnet_l1.ra2_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ecanfnet_l1_ra2-7dce93cd.pthzeca_nfnet_l2.ra3_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ecanfnet_l2_ra3-da781a61.pth)r\  r\  )rv  r~  r}  rŠ  r–  rž   )rv  r~  r}  rŠ  rƒ  znf_regnet_b1.ra2_in1kzrhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/nf_regnet_b1_256_ra2-ad85cfef.pth)r‰  rv  r~  r}  rŠ  rƒ  )r„   éð   r—  )r„   é  r˜  )r5  r5  )r„   éÈ  r™  )rv  rƒ  znf_resnet50.ra2_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/nf_resnet50_ra2-9f236009.pth)r‰  rv  r~  r}  rŠ  r  rƒ  )r0   r0   r0   gffffffî?)r„   é    rš  )r9  r9  )r‰  r�  r‚  r  r}  r~  )rk  ztest_nfnet.r160_in1kc                 ó   — t          dd| i|¤ŽS )z&NFNet-F0 (DeepMind weight compatible).r/  ro  )r/  ©ru  ©ro  r×   s     rI   r/  r/  ¹  ó   € õ ÐNÐN¸ÐNÀvÐNÐNÐNrH   c                 ó   — t          dd| i|¤ŽS )z&NFNet-F1 (DeepMind weight compatible).r1  ro  )r1  rœ  r�  s     rI   r1  r1  ¿  rž  rH   c                 ó   — t          dd| i|¤ŽS )z&NFNet-F2 (DeepMind weight compatible).r3  ro  )r3  rœ  r�  s     rI   r3  r3  Å  rž  rH   c                 ó   — t          dd| i|¤ŽS )z&NFNet-F3 (DeepMind weight compatible).r6  ro  )r6  rœ  r�  s     rI   r6  r6  Ë  rž  rH   c                 ó   — t          dd| i|¤ŽS )z&NFNet-F4 (DeepMind weight compatible).r8  ro  )r8  rœ  r�  s     rI   r8  r8  Ñ  rž  rH   c                 ó   — t          dd| i|¤ŽS )z&NFNet-F5 (DeepMind weight compatible).r=  ro  )r=  rœ  r�  s     rI   r=  r=  ×  rž  rH   c                 ó   — t          dd| i|¤ŽS )z&NFNet-F6 (DeepMind weight compatible).r?  ro  )r?  rœ  r�  s     rI   r?  r?  Ý  rž  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F0.rC  ro  )rC  rœ  r�  s     rI   rC  rC  ã  ó   € õ ÐKÐK°jÐKÀFÐKÐKÐKrH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F1.rD  ro  )rD  rœ  r�  s     rI   rD  rD  é  r¦  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F2.rE  ro  )rE  rœ  r�  s     rI   rE  rE  ï  r¦  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F3.rF  ro  )rF  rœ  r�  s     rI   rF  rF  õ  r¦  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F4.rG  ro  )rG  rœ  r�  s     rI   rG  rG  û  r¦  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F5.rH  ro  )rH  rœ  r�  s     rI   rH  rH    r¦  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F6.rI  ro  )rI  rœ  r�  s     rI   rI  rI    r¦  rH   c                 ó   — t          dd| i|¤ŽS )z	NFNet-F7.rJ  ro  )rJ  rœ  r�  s     rI   rJ  rJ    r¦  rH   c                 ó   — t          dd| i|¤ŽS )zˆNFNet-L0b w/ SiLU.

    My experimental 'light' model w/ F0 repeats, 1.5x final_conv mult, 64 group_size, .25 bottleneck & SE ratio
    rL  ro  )rL  rœ  r�  s     rI   rL  rL    s   € õ ÐKÐK°jÐKÀFÐKÐKÐKrH   c                 ó   — t          dd| i|¤ŽS )z‹ECA-NFNet-L0 w/ SiLU.

    My experimental 'light' model w/ F0 repeats, 1.5x final_conv mult, 64 group_size, .25 bottleneck & ECA attn
    rN  ro  )rN  rœ  r�  s     rI   rN  rN    ó   € õ ÐOÐO¸*ÐOÈÐOÐOÐOrH   c                 ó   — t          dd| i|¤ŽS )z‹ECA-NFNet-L1 w/ SiLU.

    My experimental 'light' model w/ F1 repeats, 2.0x final_conv mult, 64 group_size, .25 bottleneck & ECA attn
    rP  ro  )rP  rœ  r�  s     rI   rP  rP  %  r°  rH   c                 ó   — t          dd| i|¤ŽS )z‹ECA-NFNet-L2 w/ SiLU.

    My experimental 'light' model w/ F2 repeats, 2.0x final_conv mult, 64 group_size, .25 bottleneck & ECA attn
    rQ  ro  )rQ  rœ  r�  s     rI   rQ  rQ  .  r°  rH   c                 ó   — t          dd| i|¤ŽS )z‹ECA-NFNet-L3 w/ SiLU.

    My experimental 'light' model w/ F3 repeats, 2.0x final_conv mult, 64 group_size, .25 bottleneck & ECA attn
    rR  ro  )rR  rœ  r�  s     rI   rR  rR  7  r°  rH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free RegNet-B0.
    rS  ro  )rS  rœ  r�  s     rI   rS  rS  @  ó   € õ ÐOÐO¸*ÐOÈÐOÐOÐOrH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free RegNet-B1.
    rT  ro  )rT  rœ  r�  s     rI   rT  rT  G  rµ  rH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free RegNet-B2.
    rU  ro  )rU  rœ  r�  s     rI   rU  rU  N  rµ  rH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free RegNet-B3.
    rY  ro  )rY  rœ  r�  s     rI   rY  rY  U  rµ  rH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free RegNet-B4.
    r[  ro  )r[  rœ  r�  s     rI   r[  r[  \  rµ  rH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free RegNet-B5.
    r_  ro  )r_  rœ  r�  s     rI   r_  r_  c  rµ  rH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free ResNet-26.
    rb  ro  )rb  rœ  r�  s     rI   rb  rb  j  ó   € õ ÐNÐN¸ÐNÀvÐNÐNÐNrH   c                 ó   — t          dd| i|¤ŽS )z"Normalization-Free ResNet-50.
    rc  ro  )rc  rœ  r�  s     rI   rc  rc  q  r¼  rH   c                 ó   — t          dd| i|¤ŽS )z#Normalization-Free ResNet-101.
    rd  ro  )rd  rœ  r�  s     rI   rd  rd  x  rµ  rH   c                 ó   — t          dd| i|¤ŽS )zNormalization-Free SE-ResNet26.rf  ro  )rf  rœ  r�  s     rI   rf  rf    ó   € õ ÐPÐP¸:ÐPÈÐPÐPÐPrH   c                 ó   — t          dd| i|¤ŽS )zNormalization-Free SE-ResNet50.rg  ro  )rg  rœ  r�  s     rI   rg  rg  …  rÀ  rH   c                 ó   — t          dd| i|¤ŽS )z Normalization-Free SE-ResNet101.rh  ro  )rh  rœ  r�  s     rI   rh  rh  ‹  ó   € õ ÐQÐQ¸JÐQÈ&ÐQÐQÐQrH   c                 ó   — t          dd| i|¤ŽS )z Normalization-Free ECA-ResNet26.ri  ro  )ri  rœ  r�  s     rI   ri  ri  ‘  rÃ  rH   c                 ó   — t          dd| i|¤ŽS )z Normalization-Free ECA-ResNet50.rj  ro  )rj  rœ  r�  s     rI   rj  rj  —  rÃ  rH   c                 ó   — t          dd| i|¤ŽS )z!Normalization-Free ECA-ResNet101.rk  ro  )rk  rœ  r�  s     rI   rk  rk  �  s   € õ ÐRÐR¸ZÐRÈ6ÐRÐRÐRrH   c                 ó   — t          dd| i|¤ŽS )z%Test NFNet model for experimentation.rl  ro  )rl  rœ  r�  s     rI   rl  rl  £  s   € õ ÐMÐM¸
ÐMÀfÐMÐMÐMrH   )r.   )r›   NNTNN)r  NrL   NN)r  )r'  r)  r0   r,   rÃ   r%  N)r'  rÃ   TrG   rd   )r›   )erA   Úcollectionsr   Údataclassesr   r   Ú	functoolsr   Útypingr   r   r	   r
   r   r]   Útorch.nnrt   Ú	timm.datar   r   Útimm.layersr   r   r   r   r   r   r   r   r   r   r   Ú_builderr   Ú_features_fxr   Ú_manipulater   Ú	_registryr   r   Ú__all__r!   r  rK   rE   rD   re   rg   r}   rB   rF   r¶   r¿   r±   rä   r    r  r&  r,  r.  rt  ru  r†  Údefault_cfgsr/  r1  r3  r6  r8  r=  r?  rC  rD  rE  rF  rG  rH  rI  rJ  rL  rN  rP  rQ  rR  rS  rT  rU  rY  r[  r_  rb  rc  rd  rf  rg  rh  ri  rj  rk  rl  rG   rH   rI   ú<module>rÕ     sË  ððð ð$ $Ð #Ð #Ð #Ð #Ð #Ø *Ð *Ð *Ð *Ð *Ð *Ð *Ð *Ø Ð Ð Ð Ð Ð Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7à €€€Ø Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ AðJð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jð Jà *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˜'Ð
"€ð ðð ð ð ð ñ ô ñ „ðð2Eð Eð Eð Eð EˆrŒyñ Eô Eð Eñ8ð ˜Sð ¨ð ¸ð ð ð ð ð*'ð *'ð *'ð *'ð *'�B”Iñ *'ô *'ð *'ðZ ðAð Að Að Að A�B”Iñ Aô Añ ÔðAðN Ø)-Ø(,Ø#ØØñ<:ð <:Øð<:àð<:ð ð<:ð ˜XÔ&ð	<:ð
 ˜HÔ%ð<:ð ð<:ð ˆ2Œ=˜#˜t C¨ Hœ~Ð-Ô.ð<:ð <:ð <:ð <:ð@ �ØØ	ØØ	ØØ"Ø"Ø	Ø
Ø	ØØ	ØØØ	ðñ ô €ð&dð dð dð dð d�"”)ñ dô dð dðR %;Ø$(ØØ$(Ø04ñ!ð !Ø�c˜3�h”ð!à˜˜S˜”/ð!ð ˜S”Mð!ð ð	!ð
 ˜S”Mð!ð ˜d 3¨ 8œnÔ-ð!ð ð!ð !ð !ð !ñHð �u˜S #˜X”ð °%¸¸S¸´/ð Ð\að ð ð ð ð: %;ØØ!ØØØØ04ñ(ð (Ø�c˜3�h”ð(à˜˜S˜”/ð(ð ð(ð ð	(ð
 ð(ð ð(ð ð(ð ˜d 3¨ 8œnÔ-ð(ð ð(ð (ð (ð (ðZ %;ØØñ	!ð !Ø�c˜3�h”ð!à˜˜S˜”/ð!ð ð!ð ð	!ð
 ð!ð !ð !ð !ðH ˆTñ >ð >ð >à� \Ð2Ñ2Ô2Ð2ð>ð � ]Ð3Ñ3Ô3Ð3ð>ð � ]Ð3Ñ3Ô3Ð3ð	>ð
 � ^Ð4Ñ4Ô4Ð4ð>ð � _Ð5Ñ5Ô5Ð5ð>ð � _Ð5Ñ5Ô5Ð5ð>ð � _Ð5Ñ5Ô5Ð5ð>ð ˆZ˜|Ð,Ñ,Ô,Ð,ð>ð ˆZ˜}Ð-Ñ-Ô-Ð-ð>ð ˆZ˜}Ð-Ñ-Ô-Ð-ð>ð ˆZ˜~Ð.Ñ.Ô.Ð.ð>ð ˆZ˜Ð/Ñ/Ô/Ð/ð>ð  ˆZ˜Ð/Ñ/Ô/Ð/ð!>ð" ˆZ˜Ð/Ñ/Ô/Ð/ð#>ð$ ˆZ˜Ð/Ñ/Ô/Ð/ð%>ð* ˆZØ s°rÈØ�D $°1Ð5Ñ5Ô5ÀðIñ Iô Ið Ið+>ð0 �Ø s°rÈØ d d¡f¤f¸ð@ñ @ô @ð @ð1>ð6 �Ø¨°bÀtØ d d¡f¤f¸ð@ñ @ô @ð @ð7>ð< �Ø¨°bÀtØ d d¡f¤f¸ð@ñ @ô @ð @ð=>ðB �Ø¨°rÈØ d d¡f¤f¸ð@ñ @ô @ð @ðC>ðN � <Ð0Ñ0Ô0Ð0ðO>ðP � <Ð0Ñ0Ô0Ð0ðQ>ðR � <Ð:MÐNÑNÔNÐNðS>ðT � <Ð:MÐNÑNÔNÐNðU>ðV � >Ð<OÐPÑPÔPÐPðW>ðX � >Ð<OÐPÑPÔPÐPðY>ð^ �
 ,Ð/Ñ/Ô/Ð/ð_>ð` �
 ,Ð/Ñ/Ô/Ð/ða>ðb � =Ð1Ñ1Ô1Ð1ðc>ðf �* L¸TÈtÈtÐ]aÐObÑObÔObÐcÑcÔcÐcðg>ðh �* L¸TÈtÈtÐ]aÐObÑObÔObÐcÑcÔcÐcði>ðj �: ]¸tÐQUÐQUÐ_cÐQdÑQdÔQdÐeÑeÔeÐeðk>ðn �: \¸eÐQUÐQUÑQWÔQWÐXÑXÔXÐXðo>ðp �: \¸eÐQUÐQUÑQWÔQWÐXÑXÔXÐXðq>ðr �J mÀÐSWÐSWÑSYÔSYÐZÑZÔZÐZðs>ðv ˆzØÐ&7À3ÐSTÐcgØ�D $°1Ð5Ñ5Ô5ÀðIñ Iô Ið Iðw>€
ñBð  ð °$ð È#ð ÐR]ð ð ð ð ñ.ð ˆsð  3ð ¨4°°S°¬>ð ð ð ð ð( %Ð$ð e&Ø˜5˜5ØØ{Ø ]ÀMÐ\^Ðjrðtñ tô tðe&ð
 ˜5˜5ØØ{Ø ]ÀMÐ\`Ðltðvñ vô vðe&ð ˜5˜5ØØ{Ø ]ÀMÐ\`Ðltðvñ vô vðe&ð ˜5˜5ØØ{Ø }ÀmÐ^bÐnvðxñ xô xðe&ð" ˜5˜5ØØ{Ø }ÀmÐ^cÐowðyñ yô yð#e&ð* ˜5˜5ØØ{Ø }ÀmÐ^cÐowðyñ yô yð+e&ð2 ˜5˜5ØØ{Ø }ÀmÐ^cÐowðyñ yô yð3e&ð< ��Ø˜&¨]ÈMð[ñ [ô [ð=e&ð@ ��Ø˜&¨]ÈMð[ñ [ô [ðAe&ðD ��Ø˜&¨]ÈMð[ñ [ô [ðEe&ðH ��Ø˜(¨}Èmð]ñ ]ô ]ðIe&ðL ��Ø˜(¨}Èmð]ñ ]ô ]ðMe&ðP ��Ø˜(¨}Èmð]ñ ]ô ]ðQe&ðT ��Ø˜(¨}Èmð]ñ ]ô ]ðUe&ðX ��Ø˜(¨}Èmð]ñ ]ô ]ðYe&ð^ ˜˜ØØxØ ]ÀMÐadðfñ fô fð_e&ðf ˜U˜UØØ{Ø ]ÀMÐadðfñ fô fðge&ð e&ðn ˜U˜UØØ{Ø ]ÀMÐadðfñ fô fðoe&ðv ˜U˜UØØ{Ø }ÀmÐcfðhñ hô hðwe&ð~ �E�EØØ }ÀmÐcfðhñ hô hðe&ðF �E�EØ˜&¨]ÈMÐfqðsñ sô sðGe&ðJ ˜U˜UØð AØ ]ÀMÐ^iðkñ kô kðKe&ðR �E�EØ˜&¨]ÈMÐfqðsñ sô sðSe&ðV �E�EØ˜&¨]ÈMÐfqðsñ sô sðWe&ðZ �E�EØ˜(¨}ÈmÐhsðuñ uô uð[e&ð^ �E�EØ˜(¨}ÈmÐhsðuñ uô uð_e&ðd �5�5˜R¨KÐ8Ñ8Ô8ðee&ðf ˜E˜EØØ{Ø ]ÀMÐ\`Ðmxðzñ zô zðge&ðn �E�E˜b¨[Ð9Ñ9Ô9ðoe&ðr �U�U˜r¨kÐ:Ñ:Ô:ðse&ðt �U�U˜r¨kÐ:Ñ:Ô:ðue&ðv �e�e ¨{Ð;Ñ;Ô;ðwe&ðz �e�e ¨{Ð;Ñ;Ô;ð{e&ð| �e�e ¨{Ð;Ñ;Ô;ð}e&ð e&ð~ �u °Ð<Ñ<Ô<à!˜EØØ /Ø -¸6ðCñ Cô CðCe&ð e&ð e&ñ eô e€ðP ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñOð O˜Dð O°Cð O¸Kð Oð Oð Oñ „ðOð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð
 ñLð L˜ð L°ð L¸ð Lð Lð Lñ „ðLð ñPð P˜Tð P°Sð P¸[ð Pð Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pð Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pð Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pð Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pð Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pð Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pñ Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pñ Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pñ Pð Pñ „ðPð ñPð P˜Tð P°Sð P¸[ð Pñ Pð Pñ „ðPð ñOð O˜Dð O°Cð O¸Kð Oñ Oð Oñ „ðOð ñOð O˜Dð O°Cð O¸Kð Oñ Oð Oñ „ðOð ñPð P˜Tð P°Sð P¸[ð Pñ Pð Pñ „ðPð ñQð Q˜dð Q°cð Q¸kð Qñ Qð Qñ „ðQð
 ñQð Q˜dð Q°cð Q¸kð Qñ Qð Qñ „ðQð
 ñRð R˜tð R°sð R¸{ð Rñ Rð Rñ „ðRð
 ñRð R˜tð R°sð R¸{ð Rñ Rð Rñ „ðRð
 ñRð R˜tð R°sð R¸{ð Rñ Rð Rñ „ðRð
 ñSð S ð S¸ð SÀð Sñ Sð Sñ „ðSð
 ñNð N˜4ð N°3ð N¸;ð Nñ Nð Nñ „ðNð Nð NrH   