Ë
    ßGìj
6  ã                   óP  — d Z ddlZddlZddlmZ ddlmZ ddlmZ ddl	m
Z
mZmZmZmZmZmZmZmZmZ ddlmZmZmZmZ eZeZeZ e«        e«       fd„Zddd ed	«      fd
„Z ej<                  ed«      Zddd ed	«      fd„Z  ej<                  e d«      Z! ej<                  e d«      Z"ddddeefd„Z#d„ Z$ e$ejJ                  «      Z& e$e
«      Z' e$ejP                  «      Z) e$ed¬«      Z* e$ed¬«      Z+ e$e«      Z, e$e«      Z- e$e«      Z. e$e«      Z/ e$e«      Z0 e$e«      Z1d!d„Z2 e2ejf                  ejh                   «      Z5 e2ejl                  d«      Z7d„ Z8 e2ejl                  de8«      Z9d„ Z: e:«       Z:d„ Z; e;«       Z;d„ Z<d„ Z= e=«       Z=d"d„Z>d#d„Z?d„ Z@d„ ZAd „ ZBy)$a”  Stax is a small but flexible neural net specification library from scratch.

You likely do not mean to import this module! Stax is intended as an example
library only. There are a number of other much more fully-featured neural
network libraries for JAX, including `Flax`_ from Google, and `Haiku`_ from
DeepMind.

.. _Haiku: https://github.com/deepmind/dm-haiku
.. _Flax: https://github.com/google/flax
é    N)Úlax)Úrandom)
ÚreluÚlog_softmaxÚsoftmaxÚsoftplusÚsigmoidÚeluÚ
leaky_reluÚseluÚgeluÚstandardize)Úglorot_normalÚnormalÚonesÚzerosc                 ó$   ‡ ‡‡— ˆˆˆ fd„}d„ }||fS )z?Layer constructor function for a dense (fully-connected) layer.c                 ó‚   •— |d d ‰	fz   }t        j                  | «      \  }} ‰||d   ‰	f«       ‰|‰	f«      }}|||ffS )Néÿÿÿÿ)r   Úsplit)
ÚrngÚinput_shapeÚoutput_shapeÚk1Úk2ÚWÚbÚW_initÚb_initÚout_dims
          €€€ú^C:\Users\Stabi\Downloads\virtual-tryon(1)\venv\Lib\site-packages\jax/example_libraries/stax.pyÚinit_funzDense.<locals>.init_fun6   sY   ø€ Ø˜s Ð# w jÑ0€LÜ�\‰\˜#Ó�F€BˆÙ�"�{ 2‘¨Ð0Ó1±6¸"¸w¸jÓ3I€q€AØ˜!˜Q˜ÐÐó    c                 ó>   — | \  }}t        j                  ||«      |z   S ©N)ÚjnpÚdot)ÚparamsÚinputsÚkwargsr   r   s        r!   Ú	apply_funzDense.<locals>.apply_fun;   s!   € Ø�D€A€qÜ�7‰7�6˜1Ó Ñ!Ð!r#   © )r    r   r   r"   r+   s   ```  r!   ÚDenser-   4   s   ú€ ö ò
"ð 
�9Ð	Ðr#   ÚVALIDg�íµ ÷Æ°>c           
      óÜ   ‡ ‡‡‡‡‡‡‡	‡
‡‡— ‰ \  Š	ŠŠdt        ‰«      z  Š
‰xs ‰
Š‰xs* t        ‰j                  d«      ‰j                  d«      «      Šˆˆˆ ˆˆ	ˆˆˆˆˆf
d„}ˆ ˆ
ˆˆfd„}||fS )z<Layer construction function for a general convolution layer.©é   ÚIÚOc                 ó\  •
— t        ‰«      }‰D �cg c]-  }|dk(  r‰n#|dk(  r|‰j                  d«         n
t        |«      ‘Œ/ }}t        j                  ||‰‰‰«      }‰D �cg c]  }|dk(  r‰nd‘Œ }}t        j                  | «      \  }} ‰||«       ‰||«      }
}	||	|
ffS c c}w c c}w ©Nr3   r2   ÚCr1   )ÚiterÚindexÚnextr   Úconv_general_shape_tupler   r   ©r   r   Úfilter_shape_iterÚcÚkernel_shaper   Ú
bias_shaper   r   r   r   r   r   Údimension_numbersÚfilter_shapeÚlhs_specÚout_chanÚout_specÚpaddingÚrhs_specÚstridess              €€€€€€€€€€r!   r"   zGeneralConv.<locals>.init_funI   sÜ   ø€ Ü˜\Ó*Ðñ 6>ó?á5=°ð !" S¢‘HØ89¸Sº�K §¡¨sÓ 3Ò4ÜÐ*Ó+ñ,à5=ð ð ?ô ×/Ñ/Ø�\ 7¨GÐ5FóH€Lá7?Ó@±x°!˜a 3šh‘(¨AÑ-°x€JÐ@Ü�\‰\˜#Ó�F€BˆÙ�"�lÓ#¡V¨B°
Ó%;€q€AØ˜!˜Q˜ÐÐùò?ùò
 Aó   ‘2B$Á"B)c           	      óL   •— | \  }}t        j                  ||‰‰‰‰‰¬«      |z   S ©N)r@   )r   Úconv_general_dilated)	r(   r)   r*   r   r   r@   ÚonerE   rG   s	        €€€€r!   r+   zGeneralConv.<locals>.apply_funT   s9   ø€ Ø�D€A€qÜ×#Ñ# F¨A¨w¸ÀÀcØ6GôIØKLñMð Mr#   ©Úlenr   r8   )r@   rC   rA   rG   rE   r   r   r"   r+   rB   rL   rD   rF   s   ```````  @@@@r!   ÚGeneralConvrO   A   sj   ÿú€ ð "3Ñ€(ˆH�hØŒs�<Ó Ñ €#ØŠN�s€'ØÒL”] 8§>¡>°#Ó#6¸¿¹ÀsÓ8KÓL€&÷
 õ 
 ÷Mð 
�9Ð	Ðr#   )ÚNHWCÚHWIOrP   c           
      óØ   ‡ ‡‡‡‡‡‡‡
‡‡— ‰ \  Š
ŠŠdt        ‰«      z  }‰xs |Š‰xs* t        ‰j                  d«      ‰j                  d«      «      Šˆˆˆ ˆˆ
ˆˆˆˆˆf
d„}ˆ ˆˆfd„}	||	fS )zGLayer construction function for a general transposed-convolution layer.r0   r2   r3   c                 ó\  •
— t        ‰«      }‰D �cg c]-  }|dk(  r‰n#|dk(  r|‰j                  d«         n
t        |«      ‘Œ/ }}t        j                  ||‰‰‰«      }‰D �cg c]  }|dk(  r‰nd‘Œ }}t        j                  | «      \  }} ‰||«       ‰||«      }
}	||	|
ffS c c}w c c}w r5   )r7   r8   r9   r   Úconv_transpose_shape_tupler   r   r;   s              €€€€€€€€€€r!   r"   z&GeneralConvTranspose.<locals>.init_fund   sÜ   ø€ Ü˜\Ó*Ðñ 6>ó?á5=°ð !" S¢‘HØ89¸Sº�K §¡¨sÓ 3Ò4ÜÐ*Ó+ñ,à5=ð ð ?ô ×1Ñ1Ø�\ 7¨GÐ5FóH€Lá7?Ó@±x°!˜a 3šh‘(¨AÑ-°x€JÐ@Ü�\‰\˜#Ó�F€BˆÙ�"�lÓ#¡V¨B°
Ó%;€q€AØ˜!˜Q˜ÐÐùò?ùò
 ArH   c                 óH   •— | \  }}t        j                  ||‰‰‰¬«      |z   S rJ   )r   Úconv_transpose)r(   r)   r*   r   r   r@   rE   rG   s        €€€r!   r+   z'GeneralConvTranspose.<locals>.apply_funo   s5   ø€ Ø�D€A€qÜ×Ñ˜f a¨°'Ø0AôCØEFñGð Gr#   rM   )r@   rC   rA   rG   rE   r   r   rL   r"   r+   rB   rD   rF   s   ```````   @@@r!   ÚGeneralConvTransposerW   \   sj   ÿù€ ð "3Ñ€(ˆH�hØŒs�<Ó Ñ €#ØŠN�s€'ØÒL”] 8§>¡>°#Ó#6¸¿¹ÀsÓ8KÓL€&÷
 õ 
 öGð 
�9Ð	Ðr#   )ÚNHCÚHIOrX   )r   r1   é   gñhãˆµøä>Tc                 ó„   ‡ ‡‡‡‡‡‡‡	— ˆˆfd„Šˆˆfd„Š	t        j                  ‰ «      r‰ fn‰ Š ˆˆ	ˆ fd„}ˆ ˆˆˆfd„}||fS )z<Layer construction function for a batch normalization layer.c                 ó   •— ‰r	 ‰| |«      S dS ©Nr,   r,   )r   ÚshapeÚ	beta_initÚcenters     €€r!   Ú<lambda>zBatchNorm.<locals>.<lambda>|   s   ø€ ¹6¡)¨C°Ó"7Ð"IÀrÐ"Ir#   c                 ó   •— ‰r	 ‰| |«      S dS r]   r,   )r   r^   Ú
gamma_initÚscales     €€r!   ra   zBatchNorm.<locals>.<lambda>}   s   ø€ ¹U¡:¨c°5Ó#9Ð#JÈÐ#Jr#   c                 óž   •— t        ˆ	fd„t        |«      D «       «      }t        j                  | «      \  }} ‰||«       ‰||«      }}|||ffS )Nc              3   ó2   •K  — | ]  \  }}|‰vsŒ|–— Œ y ­wr%   r,   )Ú.0ÚiÚdÚaxiss      €r!   Ú	<genexpr>z.BatchNorm.<locals>.init_fun.<locals>.<genexpr>€   s   øè ø€ ÐHÑ 6™˜˜1¸!À4º-”!Ñ 6ùs   ƒ�)ÚtupleÚ	enumerater   r   )
r   r   r^   r   r   ÚbetaÚgammaÚ
_beta_initÚ_gamma_initrj   s
          €€€r!   r"   zBatchNorm.<locals>.init_fun   sP   ø€ ÜÓH¤	¨+Ô 6ÓHÓH€EÜ�\‰\˜#Ó�F€BˆÙ˜R Ó'©°R¸Ó)?ˆ%€DØ˜˜u˜Ð%Ð%r#   c                 óÚ   •— | \  }}t        ˆfd„t        t        j                  |«      «      D «       «      }t	        |‰‰	¬«      }‰r‰
r||   |z  ||   z   S ‰r|||   z   S ‰
r||   |z  S |S )Nc              3   ó@   •K  — | ]  }|‰v rd n
t        d «      –— Œ y ­wr%   )Úslice)rg   rh   rj   s     €r!   rk   z/BatchNorm.<locals>.apply_fun.<locals>.<genexpr>ˆ   s#   øè ø€ ÐNÑ;M°a�q˜D‘y‰t¤e¨D£kÓ1Ñ;Mùs   ƒ)Úepsilon)rl   Úranger&   Úndimr   )r(   Úxr*   rn   ro   ÚedÚzrj   r`   ru   rd   s          €€€€r!   r+   zBatchNorm.<locals>.apply_fun„   sy   ø€ Ø�K€Dˆ%ô 
ÓN¼5ÄÇÁÈ!ÃÔ;MÓNÓ	N€BÜ�A�t WÔ-€AÙ‘%  b¡	¨A¡°°R±Ñ 8Ð8Ù�a˜$˜r™(‘lÐ"Ù�U˜2‘Y ‘]Ð"Ø€Hr#   )r&   Úisscalar)
rj   ru   r`   rd   r_   rc   r"   r+   rp   rq   s
   ``````  @@r!   Ú	BatchNormr|   y   s;   ÿ€ ô J€*ÜJ€+Ü—L‘L Ô&ˆ$‰¨D€$ö&÷
	ð 
�9Ð	Ðr#   c                 ó    ‡ ‡— d„ }ˆ ˆfd„}||fS )z?Layer that applies a scalar function elementwise on its inputs.c                 ó
   — |dfS r]   r,   ©r   r   s     r!   ra   zelementwise.<locals>.<lambda>“   ó	   €  {°BÑ&7r#   c                 ó   •—  ‰|fi ‰¤ŽS r%   r,   )r(   r)   r*   ÚfunÚ
fun_kwargss      €€r!   ra   zelementwise.<locals>.<lambda>”   s   ø€ ©s°6Ñ/H¸ZÒ/Hr#   r,   )r‚   rƒ   r"   r+   s   ``  r!   Úelementwiser„   ‘   s   ù€ á7€(ÜH€)Ø	�9Ð	Ðr#   r   )rj   c                 ó   ‡ ‡‡— dˆˆ ˆfd„	}|S )Nc                 ó<  •‡ ‡‡‡‡— ‰xs dt        ‰ «      z  Š‰r
 ‰‰ ‰‰«      ndŠ‰€dt        ‰ «      dz   f}n"‰j                  d«      ‰j                  d«      f}t        |«      D ]  }‰ d| dz   ‰ |d z   Š ‰d| dz   ‰|d z   ŠŒ  ˆˆˆ fd„}ˆ	ˆˆ
ˆˆˆˆ fd„}||fS )	z0Layer construction function for a pooling layer.r0   Nr   r1   ÚNr6   c                 óŒ   •— t        j                  |‰‰‰«      }dt        ‰«      z  }t        j                  |‰‰|||«      }|dfS )Nr0   r,   )r   Úpadtype_to_padsrN   Úreduce_window_shape_tuple)r   r   Úpadding_valsr   Ú	out_shaperE   rG   Úwindow_shapes        €€€r!   r"   z6_pooling_layer.<locals>.PoolingLayer.<locals>.init_fun²   sS   ø€ Ü×(Ñ(¨°lØ)0°'ó;€là”C˜Ó%Ñ%€dÜ×/Ñ/Ø�\ 7¨L¸$ÀóF€ià˜ˆ]Ðr#   c                 óT   •— t        j                  |‰‰‰
‰	‰«      }‰r
 ‰||‰«      S |S r%   )r   Úreduce_window)r(   r)   r*   ÚoutÚinit_valrE   ÚreducerÚrescaleÚspecrG   r�   s       €€€€€€€r!   r+   z7_pooling_layer.<locals>.PoolingLayer.<locals>.apply_fun¹   s6   ø€ Ü×Ñ˜f h°¸Ø% wó0€cá+2‰W�S˜& $Ó'Ð;¸Ð;r#   )rN   r8   Úsorted)r�   rG   rE   r”   Únon_spatial_axesrh   r"   r+   r“   r‘   r’   Úrescalers   ````    @€€€r!   ÚPoolingLayerz$_pooling_layer.<locals>.PoolingLayer¤   sÂ   ý€ àÒ1˜¤ LÓ 1Ñ1€GÙ:B‰h�| W¨gÔ6È€Gà€|ØœC Ó-°Ñ1Ð1ÑàŸ™ C›¨$¯*©*°S«/Ð9ÐäÐ$Ö%ˆØ! " 1Ð%¨Ñ,¨|¸A¸BÐ/?Ñ?€lØ˜˜�˜dÑ" W¨Q¨R [Ñ0�gð &ö÷<ò <ð �YÐÐr#   )Nr.   Nr,   )r’   r‘   r—   r˜   s   ``` r!   Ú_pooling_layerr™   £   s   ú€ ÷ð4 
Ðr#   ç        c                 ó   ‡ ‡‡— ˆ ˆˆfd„}|S )Nc                 ó¬  •‡‡— |€d‰j                   dz
  fŠn"|j                  d«      |j                  d«      fŠt        ˆˆfd„t        ‰j                   «      D «       «      }t	        j
                  |‰j                  ¬«      }t        j                  |dt        j                  ‰‰
‰	«      }t        ‰«      D ]  }t	        j                  ||«      }Œ | |z  S )Nr   r1   r‡   r6   c              3   óD   •K  — | ]  }|‰vr‰j                   |   –— Œ y ­wr%   ©r^   )rg   rh   r)   r–   s     €€r!   rk   z=_normalize_by_window_size.<locals>.rescale.<locals>.<genexpr>Ê   s*   øè ø€ ð 8Ù#5˜aØÐ&6Ñ6ð !Ÿ,™, q�/Ù#5ùs   ƒ )Údtyperš   )rw   r8   rl   rv   r&   r   rŸ   r   r�   Úaddr•   Úexpand_dims)Úoutputsr)   r”   Úspatial_shaperL   Úwindow_sizesrh   r–   ÚdimsrE   rG   s    `     @€€€r!   r“   z*_normalize_by_window_size.<locals>.rescaleÄ   s·   ú€ Ø€|Ø˜FŸK™K¨!™OÐ+ÑàŸ™ C›¨$¯*©*°S«/Ð9Ðäô 8Ü#(¨¯©Ô#5ó8ó 8€Mô �(‰(�=¨¯©Ô
5€CÜ×$Ñ$ S¨"¬c¯g©g°t¸WÀgÓN€LÜÐ$Ö%ˆÜ—_‘_ \°1Ó5�lð &ð �\Ñ!Ð!r#   r,   )r¥   rG   rE   r“   s   ``` r!   Ú_normalize_by_window_sizer¦   Ã   s   ú€ ö"ð 
€.r#   c                  ó   — d„ } d„ }| |fS )zCLayer construction function for flattening all but the leading dim.c                 óD   — |d   t        j                  |dd  «      f}|dfS )Nr   r1   r,   )ÚmathÚprod)r   r   r   s      r!   r"   zFlatten.<locals>.init_funÙ   s*   € Ø˜q‘>¤4§9¡9¨[¸¸¨_Ó#=Ð=€LØ˜ÐÐr#   c                 óL   — t        j                  ||j                  d   df«      S )Nr   r   )r&   Úreshaper^   ©r(   r)   r*   s      r!   r+   zFlatten.<locals>.apply_funÜ   s    € Ü�;‰;�v §¡¨Q¡°Ð4Ó5Ð5r#   r,   ©r"   r+   s     r!   ÚFlattenr¯   ×   s   € òò6à	�9Ð	Ðr#   c                  ó   — d„ } d„ }| |fS )z2Layer construction function for an identity layer.c                 ó
   — |dfS r]   r,   r   s     r!   ra   zIdentity.<locals>.<lambda>ä   r€   r#   c                 ó   — |S r%   r,   r­   s      r!   ra   zIdentity.<locals>.<lambda>å   s   € ©vr#   r,   r®   s     r!   ÚIdentityr³   â   s   € á7€(Ù5€)Ø	�9Ð	Ðr#   c                 ó    ‡ — ˆ fd„}ˆ fd„}||fS )z0Layer construction function for a fan-out layer.c                 ó   •— |g‰z  dfS r]   r,   )r   r   Únums     €r!   ra   zFanOut.<locals>.<lambda>ì   s   ø€ ¨ }°sÑ':¸BÑ&?r#   c                 ó   •— |g‰z  S r%   r,   )r(   r)   r*   r¶   s      €r!   ra   zFanOut.<locals>.<lambda>í   s   ø€ °¨x¸#ª~r#   r,   )r¶   r"   r+   s   `  r!   ÚFanOutr¸   ê   s   ø€ ã?€(Û=€)Ø	�9Ð	Ðr#   c                  ó   — d„ } d„ }| |fS )z3Layer construction function for a fan-in sum layer.c                 ó   — |d   dfS )Nr   r,   r,   r   s     r!   ra   zFanInSum.<locals>.<lambda>ó   s   €  {°1¡~°rÑ&:r#   c                 ó   — t        |«      S r%   )Úsumr­   s      r!   ra   zFanInSum.<locals>.<lambda>ô   s   € ¬s°6¬{r#   r,   r®   s     r!   ÚFanInSumr½   ñ   s   € á:€(Ù:€)Ø	�9Ð	Ðr#   c                 ó    ‡ — ˆ fd„}ˆ fd„}||fS )z=Layer construction function for a fan-in concatenation layer.c                 óˆ   •‡— ‰t        |d   «      z  Št        ˆfd„|D «       «      }|d   d ‰ |fz   |d   ‰dz   d  z   }|dfS )Nr   c              3   ó(   •K  — | ]	  }|‰   –— Œ y ­wr%   r,   )rg   r^   Úaxs     €r!   rk   z0FanInConcat.<locals>.init_fun.<locals>.<genexpr>ý   s   øè ø€ Ð9©[ E�e˜B•i©[ùs   ƒr1   r,   )rN   r¼   )r   r   Úconcat_sizerŒ   rÁ   rj   s       @€r!   r"   zFanInConcat.<locals>.init_funû   s[   ù€ Ø	”�K ‘NÓ#Ñ	#€BÜÓ9©[Ó9Ó9€KØ˜A‘˜s Ð# { nÑ4°{À1±~ÀbÈÁdÀeÐ7LÑL€IØ�bˆ=Ðr#   c                 ó0   •— t        j                  |‰«      S r%   )r&   Úconcatenate)r(   r)   r*   rj   s      €r!   r+   zFanInConcat.<locals>.apply_fun   s   ø€ Ü�?‰?˜6 4Ó(Ð(r#   r,   )rj   r"   r+   s   `  r!   ÚFanInConcatrÅ   ù   s   ø€ ôô
)à	�9Ð	Ðr#   c                 ó    ‡ ‡— d„ }ˆˆ fd„}||fS )z@Layer construction function for a dropout layer with given rate.c                 ó
   — |dfS r]   r,   r   s     r!   r"   zDropout.<locals>.init_fun  s   € Ø˜ˆ?Ðr#   c                 óÊ   •— |j                  dd «      }|€d}t        |«      ‚‰dk(  r;t        j                  |‰|j                  «      }t        j                  ||‰z  d«      S |S )Nr   zÞDropout layer requires apply_fun to be called with a PRNG key argument. That is, instead of `apply_fun(params, inputs)`, call it like `apply_fun(params, inputs, rng)` where `rng` is a PRNG key (e.g. from `jax.random.key`).Útrainr   )ÚgetÚ
ValueErrorr   Ú	bernoullir^   r&   Úwhere)r(   r)   r*   r   ÚmsgÚkeepÚmodeÚrates         €€r!   r+   zDropout.<locals>.apply_fun	  sf   ø€ Ø
�*‰*�U˜DÓ
!€CØ
€{ð6€cô �s‹OÐØˆw‚Ü×Ñ˜c 4¨¯©Ó6€dÜ�Y‰Y�t˜V d™]¨AÓ.Ð.à€mr#   r,   )rÑ   rÐ   r"   r+   s   ``  r!   ÚDropoutrÒ     s   ù€ òõð 
�9Ð	Ðr#   c                  óR   ‡‡‡— t        | «      Št        | Ž \  ŠŠˆfd„}ˆˆfd„}||fS )a
  Combinator for composing layers in serial.

  Args:
    *layers: a sequence of layers, each an (init_fun, apply_fun) pair.

  Returns:
    A new layer, meaning an (init_fun, apply_fun) pair, representing the serial
    composition of the given sequence of layers.
  c                 óˆ   •— g }‰D ]7  }t        j                  | «      \  } } |||«      \  }}|j                  |«       Œ9 ||fS r%   )r   r   Úappend)r   r   r(   r"   Ú	layer_rngÚparamÚ	init_funss         €r!   r"   zserial.<locals>.init_fun(  sN   ø€ Ø€FÛˆÜ—|‘| CÓ(�n€cˆ9Ù# I¨{Ó;Ñ€k�5Ø‡m�m�EÕð ð ˜ÐÐr#   c                 óª   •— |j                  dd «      }|�t        j                  |‰«      nd‰z  }t        ‰| |«      D ]  \  }}} |||fd|i|¤Ž}Œ |S ©Nr   r%   ©Úpopr   r   Úzip)	r(   r)   r*   r   Úrngsr‚   r×   Ú
apply_funsÚnlayerss	          €€r!   r+   zserial.<locals>.apply_fun/  sb   ø€ Ø
�*‰*�U˜DÓ
!€CØ),¨Œ6�<‰<˜˜WÔ%¸gÈÑ>O€DÜ˜z¨6°4Ö8‰ˆˆU�CÙ�5˜&Ñ4 cÐ4¨VÑ4�fð 9à€Mr#   ©rN   rÝ   ©Úlayersr"   r+   rß   rØ   rà   s      @@@r!   Úserialrä     s3   ú€ ô �‹K€'Ü˜v˜,Ñ€)ˆZôõð 
�9Ð	Ðr#   c                  óT   ‡‡‡— t        | «      Št        | Ž \  ŠŠˆˆfd„}ˆˆfd„}||fS )a  Combinator for composing layers in parallel.

  The layer resulting from this combinator is often used with the FanOut and
  FanInSum layers.

  Args:
    *layers: a sequence of layers, each an (init_fun, apply_fun) pair.

  Returns:
    A new layer, meaning an (init_fun, apply_fun) pair, representing the
    parallel composition of the given sequence of layers. In particular, the
    returned layer takes a sequence of inputs and returns a sequence of outputs
    with the same length as the argument `layers`.
  c                 óž   •— t        j                  | ‰«      }t        t        ‰||«      D �� �cg c]  \  }} } || |«      ‘Œ c}} }Ž S c c}} }w r%   )r   r   rÝ   )r   r   rÞ   Úinitr^   rØ   rà   s        €€r!   r"   zparallel.<locals>.init_funI  sV   ø€ Ü�<‰<˜˜WÓ%€DÜÜ˜	 4¨Ô5õ7Ù5ñ '7 d¨C°‘�c˜5Õ!Ø5ó7ð 8ð 8ùô 7s   ®Ac                 óÔ   •— |j                  dd «      }|�t        j                  |‰
«      nd‰
z  }t        ‰	| ||«      D ����cg c]  \  }}}} |||fd|i|¤Ž‘Œ c}}}}S c c}}}}w rÚ   rÛ   )r(   r)   r*   r   rÞ   ÚfÚprx   Úrrß   rà   s            €€r!   r+   zparallel.<locals>.apply_funM  sm   ø€ Ø
�*‰*�U˜DÓ
!€CØ),¨Œ6�<‰<˜˜WÔ%¸gÈÑ>O€DÜ7:¸:ÀvÈvÐW[Ô7\Ö]Ñ7\©¨¨A¨q°!‰Aˆa�Ñ$˜Ð$˜VÓ$Ð7\Ô]Ð]ùÕ]s   ÁA"
rá   râ   s      @@@r!   Úparallelrì   8  s4   ú€ ô �‹K€'Ü˜v˜,Ñ€)ˆZõ8õ^ð 
�9Ð	Ðr#   c                 ó    ‡ — ˆ fd„}ˆ fd„}||fS )a¹  Combinator to delay layer constructor pair until input shapes are known.

  Args:
    make_layer: a one-argument function that takes an input shape as an argument
      (a tuple of positive integers) and returns an (init_fun, apply_fun) pair.

  Returns:
    A new layer, meaning an (init_fun, apply_fun) pair, representing the same
    layer as returned by `make_layer` but with its construction delayed until
    input shapes are known.
  c                 ó(   •—   ‰|«      d   | |«      S )Nr   r,   )r   r   Ú
make_layers     €r!   r"   z!shape_dependent.<locals>.init_fun`  s   ø€ Ø%‰:�kÓ" 1Ñ% c¨;Ó7Ð7r#   c                 ó>   •—   ‰|j                   «      d   | |fi |¤ŽS )Nr1   rž   )r(   r)   r*   rï   s      €r!   r+   z"shape_dependent.<locals>.apply_funb  s%   ø€ Ø&‰:�f—l‘lÓ# AÑ& v¨vÑ@¸Ñ@Ð@r#   r,   )rï   r"   r+   s   `  r!   Úshape_dependentrñ   T  s   ø€ ô8ôAà	�9Ð	Ðr#   r%   )r   )rÉ   )CÚ__doc__Ú	functoolsr©   Újaxr   r   Ú	jax.numpyÚnumpyr&   Újax.nnr   r   r   r   r	   r
   r   r   r   r   Újax.nn.initializersr   r   r   r   ÚglorotÚrandnÚ
logsoftmaxr-   rO   ÚpartialÚConvrW   ÚConv1DTransposeÚConvTransposer|   r„   ÚtanhÚTanhÚReluÚexpÚExpÚ
LogSoftmaxÚSoftmaxÚSoftplusÚSigmoidÚEluÚ	LeakyReluÚSeluÚGelur™   ÚmaxÚinfÚMaxPoolr    ÚSumPoolr¦   ÚAvgPoolr¯   r³   r¸   r½   rÅ   rÒ   rä   rì   rñ   r,   r#   r!   Ú<module>r     sç  ðñ	ó Û å Ý Ý ÷9÷ 9÷ 9ç BÓ Bð 
€Ø€Ø€
ñ (›/±&³(ó 
ð  g°dÙ˜d“|óð0 €y×Ñ˜Ð&>Ó?€ð "&¨w¸tÙ & t£óð0 $�)×#Ñ#Ð$8Ð:OÓP€Ø!�	×!Ñ!Ð"6Ø":ó<€ð  d°4¸tØ¨$óò0ñ
 �3—8‘8Ó€Ù�4Ó€Ù�#—'‘'Ó€Ù˜¨2Ô.€
Ù
�g BÔ
'€Ù�xÓ €Ù
�gÓ
€Ù�#Ó€Ù˜
Ó#€	Ù�4Ó€Ù�4Ó€óñ8 ˜Ÿ™ 3§7¡7 (Ó
+€Ù
˜Ÿ™ "Ó
%€òñ" ˜Ÿ™ "Ð&?Ó
@€òñ ‹)€òñ
 ‹:€òòñ
 ‹:€ó	óò.ò8ó8r#   