+
    ƒ°ñja"  ã                   óÎ   € ^ RI Ht ^ RIt^ RIHt  ! R R]P                  4      t ! R R]P                  4      t ! R R	]P                  4      t ! R
 R]P                  4      tR# )é    )Ú	lru_cacheN)Únnc                   ó‚   a a€ ] tR t^t oRtR
V3R lV 3R lllt]! ^R7      V3R lR l4       tV3R lR ltR	t	Vt
V ;t# )ÚNerfEmbeddera+  
An embedder module that combines input features with a 2D positional
encoding that mimics the Discrete Cosine Transform (DCT).

This module takes an input tensor of shape (B, P^2, C), where P is the
patch size, and enriches it with positional information before projecting
it to a new hidden size.
c                ó,   <€ V ^8„  d   QhRS[ RS[ RS[ /# )é   Úin_channelsÚhidden_size_inputÚ	max_freqs©Úint)ÚformatÚ__classdict__s   "€ÚDC:\Ai_Coding\apps\studio\ComfyUI\comfy\ldm\chroma_radiance\layers.pyÚ__annotate__ÚNerfEmbedder.__annotate__   s)   ø€ ÷ 
ñ 
áð
ñ ð
ñ ñ	
ó    c           	     ó¼   <€ \         SV `  4        W@n        W0n        W n        \
        P                  ! VP                  W^,          ,           W$VR7      4      V n        R# )a—  
Initializes the NerfEmbedder.

Args:
    in_channels (int): The number of channels in the input tensor.
    hidden_size_input (int): The desired dimension of the output embedding.
    max_freqs (int): The number of frequency components to use for both
                     the x and y dimensions of the positional encoding.
                     The total number of positional features will be max_freqs^2.
©ÚdtypeÚdeviceN)	ÚsuperÚ__init__r   r   r
   r   Ú
SequentialÚLinearÚembedder)Úselfr	   r
   r   r   r   Ú
operationsÚ	__class__s   &&&&&&&€r   r   ÚNerfEmbedder.__init__   sP   ø€ ô& 	‰ÑÔØŒ
Ø"ŒØ!2Ôô ŸšØ×Ñ˜k°q­LÕ8Ð:KÐagÐÓhó
ˆŽr   )Úmaxsizec                ón   <€ V ^8„  d   QhRS[ RS[P                  RS[P                  RS[P                  /# )r   Ú
patch_sizer   r   Úreturn)r   Útorchr   r   ÚTensor)r   r   s   "€r   r   r   0   s8   ø€ ÷ 4ñ 4¡Cð 4±·±ð 4ÁeÇkÁkð 4ÑV[×VbÑVbñ 4r   c                óÎ  € \         P                  ! ^ ^WVR7      p\         P                  ! ^ ^WVR7      p\         P                  ! WTRR7      w  rTVP                  R^^4      pVP                  R^^4      p\         P                  ! ^ V P                  ^,
          V P                  W2R7      pVR,          pVR,          p^Wx,          ,           R,          p	\         P
                  ! WG,          \         P                  ,          4      p
\         P
                  ! WX,          \         P                  ,          4      pW«,          V	,          P                  ^RV P                  ^,          4      pV# )aÙ  
Generates and caches 2D DCT-like positional embeddings for a given patch size.

The LRU cache is a performance optimization that avoids recomputing the
same positional grid on every forward pass.

Args:
    patch_size (int): The side length of the square input patch.
    device: The torch device to create the tensors on.
    dtype: The torch dtype for the tensors.

Returns:
    A tensor of shape (1, patch_size^2, max_freqs^2) containing the
    positional embeddings.
)r   r   Úij)Úindexingr   éÿÿÿÿ)NºNNNN)NNr+   )r%   ÚlinspaceÚmeshgridÚreshaper   ÚcosÚpiÚview)r   r#   r   r   Úpos_xÚpos_yÚfreqsÚfreqs_xÚfreqs_yÚcoeffsÚdct_xÚdct_yÚdcts   &&&&         r   Ú	fetch_posÚNerfEmbedder.fetch_pos/   s  € ô$ —’˜q ! ZÀeÔLˆÜ—’˜q ! ZÀeÔLˆô —~’~ e¸TÔB‰ˆð —‘˜b ! QÓ'ˆØ—‘˜b ! QÓ'ˆô —’˜q $§.¡.°1Õ"4°d·n±nÈEÔaˆð
 ˜Õ&ˆØ˜Õ&ˆð �gÕ'Õ'¨BÕ.ˆô —	’	˜%�/¬E¯H©HÕ4Ó5ˆÜ—	’	˜%�/¬E¯H©HÕ4Ó5ˆð �}˜vÕ%×+Ñ+¨A¨r°4·>±>ÀQÕ3FÓGˆàˆ
r   c                óN   <€ V ^8„  d   QhRS[ P                  RS[ P                  /# )r   Úinputsr$   ©r%   r&   )r   r   s   "€r   r   r   f   s#   ø€ ÷ ;ñ ;™eŸl™lð ;©u¯|©|ñ ;r   c                óz  € VP                   w  r#p\        VR,          4      pVP                  pVP                  V P                  R7      pV P	                  WQP
                  V P                  4      pVP                  V^^4      p\        P                  ! W3RR7      pV P                  V4      P                  VR7      # )z±
Forward pass for the embedder.

Args:
    inputs (Tensor): The input tensor of shape (B, P^2, C).

Returns:
    Tensor: The output tensor of shape (B, P^2, hidden_size_input).
g      à?)r   ©Údimr*   )
Úshaper   r   Útor;   r   Úrepeatr%   Úcatr   )r   r>   ÚBÚP2ÚCr#   Úinput_dtyper:   s   &&      r   ÚforwardÚNerfEmbedder.forwardf   sž   € ð —<‘<‰ˆˆqô ˜˜s�“^ˆ
à—l‘lˆØ—‘ §¡�Ó,ˆð �n‰n˜Z¯©¸¿
¹
ÓCˆð �j‰j˜˜A˜qÓ!ˆô —’˜F˜=¨bÔ1ˆð �}‰}˜VÓ$×'Ñ'¨kÐ'Ó:Ð:r   )r   r   r
   r   ©NNN)Ú__name__Ú
__module__Ú__qualname__Ú__firstlineno__Ú__doc__r   r   r;   rK   Ú__static_attributes__Ú__classdictcell__Ú__classcell__©r   r   s   @@r   r   r      s=   ù‡ € ñ÷
õ 
ñ< �qÔ÷4ó ð4÷l;÷ ;ð ;r   r   c                   óX   a a€ ] tR t^‡t oRtRV3R lV 3R llltV3R lR ltRtVtV ;t	# )ÚNerfGLUBlockz7
A NerfBlock using a Gated Linear Unit (GLU) like MLP.
c                ó&   <€ V ^8„  d   QhRS[ RS[ /# )r   Úhidden_size_sÚhidden_size_xr   )r   r   s   "€r   r   ÚNerfGLUBlock.__annotate__‹   s   ø€ ÷ #ñ #¡cð #¹#ñ #r   c                ó¾   <€ \         SV `  4        ^V^,          ,          V,          pVP                  WWER7      V n        VP	                  W$VR7      V n        W0n        R# )é   r   N)r   r   r   Úparam_generatorÚRMSNormÚnormÚ	mlp_ratio)	r   rZ   r[   rb   r   r   r   Útotal_paramsr   s	   &&&&&&& €r   r   ÚNerfGLUBlock.__init__‹   sY   ø€ Ü‰ÑÔð ˜=¨!Õ+Õ+¨iÕ7ˆØ)×0Ñ0°ÐTYÐ0ÓiˆÔØ×&Ñ& }È&Ð&ÓQˆŒ	Ø"Žr   c                óh   <€ V ^8„  d   QhRS[ P                  RS[ P                  RS[ P                  /# )r   ÚxÚsr$   r?   )r   r   s   "€r   r   r\   –   s.   ø€ ÷ ñ ™Ÿ™ð ©%¯,©,ð ¹5¿<¹<ñ r   c                óN  € VP                   w  r4pV P                  V4      pVP                  ^RR7      w  rxp	VP                  W5WPP                  ,          4      p
VP                  W5WPP                  ,          4      pV	P                  W5V P                  ,          V4      p\
        P                  P                  P                  V
RR7      p
\
        P                  P                  P                  VRR7      p\
        P                  P                  P                  VRR7      pTpV P                  V4      p\
        P                  ! \
        P                  P                  P                  \
        P                  ! W4      4      \
        P                  ! W4      ,          V4      pW,           # )r^   rA   r*   éþÿÿÿ)rC   r_   Úchunkr1   rb   r%   r   Ú
functionalÚ	normalizera   ÚbmmÚsilu)r   rf   rg   Ú
batch_sizeÚnum_xr[   Ú
mlp_paramsÚfc1_gate_paramsÚfc1_value_paramsÚ
fc2_paramsÚfc1_gateÚ	fc1_valueÚfc2Úres_xs   &&&           r   rK   ÚNerfGLUBlock.forward–   sH  € Ø+,¯7©7Ñ(ˆ
˜=Ø×)Ñ)¨!Ó,ˆ
ð 9C×8HÑ8HÈÐPRÐ8HÓ8SÑ5ˆ¨:ð #×'Ñ'¨
À=×SaÑSaÕCaÓbˆØ$×)Ñ)¨*À]×UcÑUcÕEcÓdˆ	Ø�o‰o˜j¸$¿.¹.Õ*HÈ-ÓXˆô —8‘8×&Ñ&×0Ñ0°¸rÐ0ÓBˆÜ—H‘H×'Ñ'×1Ñ1°)ÀÐ1ÓDˆ	Ü�h‰h×!Ñ!×+Ñ+¨C°RÐ+Ó8ˆàˆØ�I‰I�a‹Lˆô �IŠI”e—h‘h×)Ñ)×.Ñ.¬u¯yªy¸Ó/EÓFÌÏÊÐSTÓI`Õ`ÐbeÓfˆà�yÐr   )rb   ra   r_   rM   )
rN   rO   rP   rQ   rR   r   rK   rS   rT   rU   rV   s   @@r   rX   rX   ‡   s#   ù‡ € ñ÷#õ #÷÷ ð r   rX   c                   óH   a a€ ] tR t^°t oRV 3R lltV3R lR ltRtVtV ;t# )ÚNerfFinalLayerc                ó„   <€ \         SV `  4        VP                  WVR 7      V n        VP	                  WW4R 7      V n        R# )r   N)r   r   r`   ra   r   Úlinear©r   Úhidden_sizeÚout_channelsr   r   r   r   s   &&&&&&€r   r   ÚNerfFinalLayer.__init__±   s<   ø€ Ü‰ÑÔØ×&Ñ& {ÈÐ&ÓOˆŒ	Ø ×'Ñ'¨ÈÐ'Ó^ˆŽr   c                óN   <€ V ^8„  d   QhRS[ P                  RS[ P                  /# ©r   rf   r$   r?   )r   r   s   "€r   r   ÚNerfFinalLayer.__annotate__¶   s'   ø€ ÷ Gñ G™Ÿ™ð G©%¯,©,ñ Gr   c                ó‚   € V P                  V P                  VP                  ^R4      4      4      P                  R^4      # ©é   r*   )r}   ra   Úmovedim©r   rf   s   &&r   rK   ÚNerfFinalLayer.forward¶   s4   € ð �{‰{˜4Ÿ9™9 Q§Y¡Y¨q°"Ó%5Ó6Ó7×?Ñ?ÀÀAÓFÐFr   )r}   ra   rM   ©	rN   rO   rP   rQ   r   rK   rS   rT   rU   rV   s   @@r   r{   r{   °   s   ù‡ € ÷_÷
G÷ Gð Gr   r{   c                   óT   a a€ ] tR t^¼t oRV3R lV 3R llltV3R lR ltRtVtV ;t# )ÚNerfFinalLayerConvc                ó&   <€ V ^8„  d   QhRS[ RS[ /# )r   r   r€   r   )r   r   s   "€r   r   ÚNerfFinalLayerConv.__annotate__½   s   ø€ ÷ 

ñ 

¡Cð 

±sñ 

r   c           	     óŒ   <€ \         SV `  4        VP                  WVR 7      V n        VP	                  VV^^VVR7      V n        R# )r   )r	   r€   Úkernel_sizeÚpaddingr   r   N)r   r   r`   ra   ÚConv2dÚconvr~   s   &&&&&&€r   r   ÚNerfFinalLayerConv.__init__½   sN   ø€ Ü‰ÑÔØ×&Ñ& {ÈÐ&ÓOˆŒ	Ø×%Ñ%Ø#Ø%ØØØØð &ó 
ˆŽ	r   c                óN   <€ V ^8„  d   QhRS[ P                  RS[ P                  /# rƒ   r?   )r   r   s   "€r   r   r�   É   s'   ø€ ÷ Eñ E™Ÿ™ð E©%¯,©,ñ Er   c                ó‚   € V P                  V P                  VP                  ^R4      4      P                  R^4      4      # r†   )r”   ra   rˆ   r‰   s   &&r   rK   ÚNerfFinalLayerConv.forwardÉ   s4   € ð �y‰y˜Ÿ™ 1§9¡9¨Q°Ó#3Ó4×<Ñ<¸RÀÓCÓDÐDr   )r”   ra   rM   r‹   rV   s   @@r   r�   r�   ¼   s!   ù‡ € ÷

õ 

÷E÷ Eð Er   r�   )	Ú	functoolsr   r%   r   ÚModuler   rX   r{   r�   © r   r   Ú<module>rœ      sV   ðå ã Ý ô|;�2—9‘9ô |;ô~&�2—9‘9ô &ôR	G�R—Y‘Yô 	GôE˜Ÿ™ö Er   