+h`SSKrSSKJrJrJr SSKrSSKJr SSKJ r J r SSK J r SSK JrJr SSKJrJr SS KJr SS KJr SS KJr SS KJr SS KJr "SS\R:5r"SS\R:5r"SS\R:5r "SS\R:5r!"SS\R:5r""SS\R:5r#"SS\\ 5r$Sr%g)N)OptionalTupleUnion) ConfigMixinregister_to_config)apply_forward_hook) Attention SpatialNorm) DecoderOutputDiagonalGaussianDistribution) Downsample2D)AutoencoderKLOutput) ModelMixin) ResnetBlock2D) Upsample2Dc^\rSrSrSrSS\S\\S\S\S\S \S \S S4U4S jjjr \ S \ RS \ R4Sj5r S \ RS\S \ R4SjrSrU=r$)AllegroTemporalConvLayer!z Temporal convolutional layer that can be used for video (sequence of images) input. Code adapted from: https://github.com/modelscope/modelscope/blob/1509fdb973e5871f37148a4b5e5964cafd43e64d/modelscope/models/multi_modal/video_synthesis/unet_sd.py#L1016 Nin_dimout_dimdropoutnorm_num_groups up_sample down_samplestridereturnc D>[T U]5 U=(d Un[US- S-5=pSn X`lXPlU(a^[ R "[ R"XA5[ R"5[ R"XSXw4SSX4S95Ul OU(a`[ R "[ R"XA5[ R"5[ R"XS-SXw4SX4S95Ul O\[ R "[ R"XA5[ R"5[ R"XSXw4XU 4S95Ul [ R "[ R"XB5[ R"5[ R"U5[ R"X!SXw4XU 4S95Ul [ R "[ R"XB5[ R"5[ R"U5[ R"X!SXw4XU4S95Ul [ R "[ R"XB5[ R"5[ R"X!SXw4XU4S95Ulg) Ng?rr )r r r )rpaddingr!r)super__init__intrrnn Sequential GroupNormSiLUConv3dconv1Dropoutconv2conv3conv4) selfrrrrrrrpad_hpad_wpad_t __class__s n/home/james-whalen/.local/lib/python3.13/site-packages/diffusers/models/autoencoders/autoencoder_kl_allegro.pyr$!AllegroTemporalConvLayer.__init__'s #VVaZ3.//&"  _5  &Av+>y[\^cZklDJ  _5  &A+6/BQPUL]^DJ  _5  &Av+>W\H]^DJ ]] LL 2 GGI JJw  IIg6':USXDY Z   ]] LL 2 GGI JJw  IIg6':USXDY Z   ]] LL 2 GGI IIg6':USXDY Z  hidden_statesc[R"USS2SS2SS24U4SS9n[R"XSS2SS2SS244SS9nU$)Nrr r )dim)torchcat)r8s r5_pad_temporal_dim*AllegroTemporalConvLayer._pad_temporal_dim^sN =Aqs#;]"KQRS  =1bc 2J"KQRS r7 batch_sizecXURSUS45RSSSSS5nUR(aUSS2SS2SSS24nO5UR(a"UR SSUR SS-S9nOUnUR(dUR(aUR U5nO"URU5nUR U5nUR(a6URSS5RSSSSSS 5RSS5nURU5nURU5nURU5nURU5nURU5nURU5nX1-nURSSSSS5RSS5nU$) Nrr;r r r)r: output_size)r r;) unflattenpermuterrrepeat_interleaveshaper+r>flattenr-r.r/)r0r8r@identitys r5forward AllegroTemporalConvLayer.forwardds%//J3CDLLQPQSTVWYZ[   $Q3Q3Y/H ^^$66qa]M`M`abMcfgMg6hH$H   t~~ JJ}5M 22=AM JJ}5M >>)33Aw?GG1aQRTUWXYaabcefgM..}=  =1 ..}=  =1 ..}=  =1  0 %--aAq!<DDQJ r7)r+r-r.r/rr)N FFr )__name__ __module__ __qualname____firstlineno____doc__r%rfloatboolr$ staticmethodr<Tensorr>rK__static_attributes__ __classcell__r4s@r5rr!s"&!!5 5 #5  5  5  5 5 5  5 5 n%,, U\\su||r7rc^\rSrSrSS\S\S\S\S\S\S\S \S \S \S \S \S\4U4SjjjrS\ RS\ R4Sjr Sr U=r $)AllegroDownBlock3D in_channels out_channelsr num_layers resnet_epsresnet_time_scale_shift resnet_act_fn resnet_groupsresnet_pre_normoutput_scale_factorspatial_downsampletemporal_downsampledownsample_paddingc>[TU]5 /n/n[U5HInUS:XaUOUnUR[ UUSUUUUUU U S9 5 UR[ UUSUS95 MK [ R"U5Ul[ R"U5Ul U (a[ X"SUSSS9Ul Xl U (a'[ R"[USX-SS 9/5Ul gSUl g) Nr r^r_ temb_channelsepsgroupsrtime_embedding_norm non_linearityrfpre_norm皙?rrTr)rrrrop)use_convr_r!name)r#r$rangeappendrrr& ModuleListresnets temp_convstemp_convs_downadd_temp_downsampler downsamplers)r0r^r_rr`rarbrcrdrerfrgrhrirzr{ir4s r5r$AllegroDownBlock3D.__init__s  z"A)*a+\K NN +!-"&"(#(?"/(;,     (  $1  !#2}}W- -- 3 #;Cdhqr$D $7  "  $t,qu!D !%D r7r8rcURSnURSSSSS5RSS5n[URUR 5Hup4U"USS9nU"XS9nM UR (aURXS9nURbURH nU"U5nM URSUS45RSSSSS5nU$ Nrr r rrB)temb)r@r;) rHrFrIziprzr{r}r|r~rE)r0r8r@resnet temp_conv downsamplers r5rKAllegroDownBlock3D.forwards"((+ %--aAq!<DDQJ !$T\\4??!C F"=tdefaultswishrNT?TFr rOrPrQrRr%rTstrrUr$r<rWrKrXrYrZs@r5r\r\s  '0$ $%(#'$)"#@%@%@% @%  @%  @%"%@%@%@%@%#@%!@%"@% @%@%DU\\ellr7r\c^\rSrSrSS\S\S\S\S\S\S\S \S \S \S \S \S\\4U4Sjjjr S\ RS\ R4Sjr Sr U=r$)AllegroUpBlock3Dr^r_rr`rarbrcrdrerfspatial_upsampletemporal_upsamplerlc>[TU]5 /n/n[U5HInUS:XaUOUnUR[ UUU UUUUUU U S9 5 UR[ UUSUS95 MK [ R"U5Ul[ R"U5Ul Xl U (a[ X"SUSSS9Ul U (a&[ R"[USUS9/5Ul gSUl g) NrrkrrrsTr)rrrr)rur_)r#r$rwrxrrr&ryrzr{add_temp_upsample temp_conv_upr upsamplers)r0r^r_rr`rarbrcrdrerfrrrlrzr{rinput_channelsr4s r5r$AllegroUpBlock3D.__init__s  z"A,-F[ N NN .!-"/"(#(?"/(;,     (  $1  ##4}}W- -- 3!2  8Cbfop!D   mmZ tbn-o,pqDO"DOr7r8rcURSnURSSSSS5RSS5n[URUR 5Hup4U"USS9nU"XS9nM UR (aURXS9nURbURH nU"U5nM URSUS45RSSSSS5nU$r) rHrFrIrrzr{rrrrE)r0r8r@rr upsamplers r5rKAllegroUpBlock3D.forwards"((+ %--aAq!<DDQJ !$T\\4??!C F"=t[TU]5 [UUUUUUUUU U S9 /n [UUSUS9/n/nU cUn [ U5HnU (a5UR [ UX-U U UUS:XaUOSUS:XaUOSSSSSS9 5 OUR S5 U R [UUUUUUUUU U S9 5 UR [UUSUS95 M [R"U 5Ul [R"U5Ul [R"U5Ul g)NrkrrrsrspatialT) headsdim_headrescale_output_factorrmrspatial_norm_dimresidual_connectionbiasupcast_softmax_from_deprecated_attn_block) r#r$rrrwrxr r&ryrzr{ attentions)r0r^rlrr`rarbrcrdrerrrfrzr{r_r4s r5r$AllegroMidBlock3DConv.__init__0sk  '(+$$;+$7(    % -     %!, z"A!!#)?!3.A&9PT]9] cg:QU^:^dh,0!'+48  !!$' NN +!,"/"(#(?"/(;,     ($1  G#X}}W- -- 3-- 3r7r8rcURSnURSSSSS5RSS5nURS"USS9nURS"XS9n[ UR URSSURSS5Hup4nU"U5nU"USS9nU"XS9nM URSUS45RSSSSS5nU$r)rHrFrIrzr{rrrE)r0r8r@attnrrs r5rKAllegroMidBlock3DConv.forwards"((+ %--aAq!<DDQJ  Q DA *=P '*4??DLL[TU]5 [R"UUSSSSS9Ul[R "USUSSSS9Ul[R"/5UlUSn [U5HaupU n XKn U [U5S- :HnU S:Xa[UU U U(+X[S SUUS 9 nO [S 5eURRU5 Mc [US S USS US USS9Ul[R "US US S9Ul[R$"5UlU (aSU-OUn[R "US US SSS9Ul[R"US USSS9UlSUlg)Nrrr  kernel_sizerr!rr r r rr)r^r_rr!r\r) r`r^r_rgrhrarircrdzCInvalid `down_block_type` encountered. Must be `AllegroDownBlock3D`r;rr^rarcrfrbrrdrl num_channels num_groupsrmr r"F)r#r$r&Conv2dconv_inr* temp_conv_inry down_blocks enumeratelenr\ ValueErrorrxr mid_blockr( conv_norm_outr)conv_act temp_conv_outconv_outgradient_checkpointing)r0r^r_rrrrrrroutput_channelrdown_block_type input_channelis_final_block down_blockconv_out_channelsr4s r5r$AllegroEncoder3D.__init__s" yy  q !   II*1-+A.!  ==,,A."+,<"= A*M/2N#&8"9A"==N"66// -!/+9'9(B(E#'("("1  !!fgg    # #J /)#>./*2. !$-1"5)   \\7I"7MZiost  08A ,lYY'9"'=?QRT?UW`jst "4R"8:KQXYZ &+#r7samplercURSnURSSSSS5RSS5nURU5nUR SUS45RSSSSS5nUnUR U5nX-n[ R"5(aRUR(aAURHnURXA5nM URURU5nO,URH nU"U5nM URU5nURSSSSS5RSS5nURU5nURU5nUR SUS45RSSSSS5nUnURU5nX-nURSSSSS5RSS5nURU5nUR SUS45RSSSSS5nU$Nrr r rrBr;)rHrFrIrrErr<is_grad_enabledrr_gradient_checkpointing_funcrrrrr)r0rr@residualrs r5rKAllegroEncoder3D.forwards\\!_ 1aA.66q!<f%!!!j"%56>>q!Q1M""6*"  " "t'B'B".. :::N/66t~~vNF#.. #F+/^^F+F1aA.66q!<##F+v&!!!j"%56>>q!Q1M##F+"1aA.66q!<v&!!!j"%56>>q!Q1M r7) rrrrrrrrrrOrPrQrRr%rrrUr$r<rWrKrXrYrZs@r5rrs- /C7Q !!Q,Q,Q, S/ Q,"#s(OQ,%*$)$4Q,Q,Q,Q,Q,Q,f(ell(u||((r7rc^\rSrSrSSS/SQSSSS S 4 S \S \S \\S4S\\S4S\\S4S\S\S\S\4U4SjjjrS\ RS\ R4Sjr Sr U=r $)AllegroDecoder3DirBrrrrrFTTFrr rNrgroupr^r_up_block_types.temporal_upsample_blocksrrrr norm_typec >[TU]5 [R"UUSSSSS9Ul[R "USUSSSS9UlSUl[R"/5Ul U S:XaUOSn [USS USU S :XaS OU USUU S 9Ul[[U55n U S n [U5HgupU nXn U [U5S- :HnUS:Xa[US-UU U(+XMS UUU U S9 nO [!S5eURR#U5 U nMi U S:Xa[%US U 5UlO[R("US US S9Ul[R*"5Ul[R "US US SSS9Ul[R"US USSS9UlSUlg)Nr;rr rrrr"rrrrrrr) r`r^r_rrrarcrdrlrbz?Invalid `UP_block_type` encountered. Must be `AllegroUpBlock3D`rF)r#r$r&rrr*rrry up_blocksrlistreversedrrrrrxr rr(r)rrrr)r0r^r_rrrrrrrrlreversed_block_out_channelsrr up_block_typeprev_output_channelrup_blockr4s r5r$AllegroDecoder3D.__init__s" yy  r "   II&8&<>PQS>TV_irsr*'0I'= 4 /*2. !1:g1EI91"5)'  '+84F+G&H#4Q7 ). 9 A"0 8;N#&8"9A"==N 22+/!3 3!/)7%7&>&A#"("1"/,5 !!bcc NN ! !( +"0 /!:4  !!,-?-BM!RD !#;Ma;P]lrv!wD   YY'9!'<>PQR>SU^hqr "4Q"7qRST &+#r7rrcURSnURSSSSS5RSS5nURU5nUR SUS45RSSSSS5nUnUR U5nX-n[ [URR555Rn[R"5(aRUR(aAURURU5nURHnURXQ5nM O=URU5nUR!U5nURH nU"U5nM URSSSSS5RSS5nUR#U5nUR%U5nUR SUS45RSSSSS5nUnUR'U5nX-nURSSSSS5RSS5nUR)U5nUR SUS45RSSSSS5nU$r)rHrFrIrrErnextiterr parametersdtyper<rrrrtorrrr)r0rr@r upscale_dtypers r5rKAllegroDecoder3D.forwardus\\!_ 1aA.66q!<f%!!!j"%56>>q!Q1M""6*"T$..";";"=>?EE  " "t'B'B66t~~vNF!NN::8L+ ^^F+FYY}-F!NN!&)+1aA.66q!<##F+v&!!!j"%56>>q!Q1M##F+"1aA.66q!<v&!!!j"%56>>q!Q1M r7) rrrrrrrrrrrZs@r5rrs+ 6P.B !! U,U,U,c3h U,#(c "2U,"#s(OU,U,U,U,U,U,n,ell,u||,,r7rc"^\rSrSrSrSr\S)S\S\S\\ S4S\\ S4S \\S4S \\ S4S \\ S4S \S \S\ S\S\ S\S\ S\ SS4 U4Sjjj5r S*Sjr S*SjrS*SjrS*SjrS\R$S\R$4Sjr\S+S\R$S\ S\\\\44Sjj5rS\R$S\R$4S jr\S+S\R$S\ S\\\R$44S!jj5rS\R$S\R$4S"jrS\R$S\R$4S#jrS,S$\R$S%\ S\ S&\\R>S\\\R$44 S'jjr S(r!U=r"$)-AutoencoderKLAllegroia} A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used in [Allegro](https://github.com/rhymes-ai/Allegro). This model inherits from [`ModelMixin`]. Check the superclass documentation for it's generic methods implemented for all models (such as downloading or saving). Parameters: in_channels (int, defaults to `3`): Number of channels in the input image. out_channels (int, defaults to `3`): Number of channels in the output. down_block_types (`Tuple[str, ...]`, defaults to `("AllegroDownBlock3D", "AllegroDownBlock3D", "AllegroDownBlock3D", "AllegroDownBlock3D")`): Tuple of strings denoting which types of down blocks to use. up_block_types (`Tuple[str, ...]`, defaults to `("AllegroUpBlock3D", "AllegroUpBlock3D", "AllegroUpBlock3D", "AllegroUpBlock3D")`): Tuple of strings denoting which types of up blocks to use. block_out_channels (`Tuple[int, ...]`, defaults to `(128, 256, 512, 512)`): Tuple of integers denoting number of output channels in each block. temporal_downsample_blocks (`Tuple[bool, ...]`, defaults to `(True, True, False, False)`): Tuple of booleans denoting which blocks to enable temporal downsampling in. latent_channels (`int`, defaults to `4`): Number of channels in latents. layers_per_block (`int`, defaults to `2`): Number of resnet or attention or temporal convolution layers per down/up block. act_fn (`str`, defaults to `"silu"`): The activation function to use. norm_num_groups (`int`, defaults to `32`): Number of groups to use in normalization layers. temporal_compression_ratio (`int`, defaults to `4`): Ratio by which temporal dimension of samples are compressed. sample_size (`int`, defaults to `320`): Default latent size. scaling_factor (`float`, defaults to `0.13235`): The component-wise standard deviation of the trained latent space computed using the first batch of the training set. This is used to scale the latent space to have unit variance when training the diffusion model. The latents are scaled with the formula `z = z * scaling_factor` before being passed to the diffusion model. When decoding, the latents are scaled back to the original scale with the formula: `z = 1 / scaling_factor * z`. For more details, refer to sections 4.3.2 and D.1 of the [High-Resolution Image Synthesis with Latent Diffusion Models](https://huggingface.co/papers/2112.10752) paper. force_upcast (`bool`, default to `True`): If enabled it will force the VAE to run in float32 for high image resolution pipelines, such as SD-XL. VAE can be fine-tuned / trained to a lower range without losing too much precision in which case `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix Tr^r_r.rrrrlatent_channelsrrrtemporal_compression_ratio sample_sizescaling_factor force_upcastrNc >[TU]5 [UUUUUU U U SS9 Ul[ UUUUUU U U S9Ul[ R"SU-SU-S5Ul[ R"XS5Ul SUl SUl S[U5S- -Ul SUlSUlS UlS nUX4UlUUR- XR- XR - 4Ulg) NT) r^r_rrrrrrr)r^r_rrrrrrr r FxP)r#r$rencoderrdecoderr&r quant_convpost_quant_conv use_slicing use_tilingrspatial_compression_ratiotile_overlap_ttile_overlap_htile_overlap_wkernelr)r0r^r_rrrrrrrrrrrrr sample_framesr4s r5r$AutoencoderKLAllegro.__init__s: '#(-'A1-+   ('%)%=1-+   ))A$7_9LaP!yy1M !)*s3E/F/J)K&!  $k? D// / -- - -- -  r7cSUlg)z Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow processing larger images. TNrr0s r5 enable_tiling"AutoencoderKLAllegro.enable_tilings r7cSUlg)z Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing decoding in one step. FNrrs r5disable_tiling#AutoencoderKLAllegro.disable_tiling&s  r7cSUlg)z Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes. TNrrs r5enable_slicing#AutoencoderKLAllegro.enable_slicing-s  r7cSUlg)z Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing decoding in one step. FNr rs r5disable_slicing$AutoencoderKLAllegro.disable_slicing4s !r7xc\UR(aURU5$[S5e)Nz5Encoding without tiling has not been implemented yet.)r tiled_encodeNotImplementedError)r0r&s r5_encodeAutoencoderKLAllegro._encode;( ??$$Q' '!"YZZr7 return_dictc>UR(aYURSS:aFURS5Vs/sHo0RU5PM nn[R "U5nOURU5n[ U5nU(dU4$[US9$s snf)a Encode a batch of videos into latents. Args: x (`torch.Tensor`): Input batch of videos. return_dict (`bool`, defaults to `True`): Whether to return a [`~models.autoencoder_kl.AutoencoderKLOutput`] instead of a plain tuple. Returns: The latent representations of the encoded videos. If `return_dict` is True, a [`~models.autoencoder_kl.AutoencoderKLOutput`] is returned, otherwise a plain `tuple` is returned. rr ) latent_dist)rrHsplitr*r<r=rr)r0r&r-x_sliceencoded_slicesh posteriors r5encodeAutoencoderKLAllegro.encodeCs"    QCD771:N:ll73:NN .)A QA03 < "y99OsBzc\UR(aURU5$[S5e)Nz5Decoding without tiling has not been implemented yet.)r tiled_decoder))r0r7s r5_decodeAutoencoderKLAllegro._decode`r,r7c(UR(aYURSS:aFURS5Vs/sHo0RU5PM nn[R "U5nOURU5nU(dU4$[ US9$s snf)a Decode a batch of videos. Args: z (`torch.Tensor`): Input batch of latent vectors. return_dict (`bool`, defaults to `True`): Whether to return a [`~models.vae.DecoderOutput`] instead of a plain tuple. Returns: [`~models.vae.DecoderOutput`] or `tuple`: If return_dict is True, a [`~models.vae.DecoderOutput`] is returned, otherwise a plain `tuple` is returned. rr r)rrHr0r:r<r=r )r0r7r-z_slicedecoded_slicesdecodeds r5decodeAutoencoderKLAllegro.decodehsu    QCD771:N:ll73:NNii/Gll1oG: G,,OsBc  SnURnURRnURupVpxn [R "XpR S- URS- 5S-n [R "XR S- URS- 5S-n [R "XR S- URS- 5S-n Sn URX-U -SURR-UR SU-UR SU-UR SU-45nURX&UR SUR SUR S45n[U 5GHn[U 5GHrn[U 5GH^nUURS-UURS-UR S-nnUURS-UURS-UR S-nnUURS-UURS-UR S-nnUSS2SS2UU2UU2UU24nUXU-'X-US- :XdXU -U -S- :XaURU5nXU -U -S- :XaX-US- :waUSX-S-XX-- S&O UXU- S-U S-&URX&UR SUR SUR S45nU S- n GMa GMu GM URUSURR-Xt-X-X-45nUR SU-UR SU-UR SU-4nURSU-URSU-URSU-4nUSUS- USUS- USUS- 4n[U 5HnUUS-UUS-US-nn[U 5HnUUS-UUS-US-nn[U 5HvnUUS-UUS-US-nn[UU US4UU US4UU US4UUU -U -UU --U-RS55nUSS2SS2UU2UU2UU24==U- ss'Mx M M URSSSSS5RSS5nUR!U5nUR#SUS45RSSSSS5nU$Nr rr rrBr;)rconfigrrHmathfloorrr new_zerosrrwr _prepare_for_blend unsqueezerFrIr rE)r0r&local_batch_sizersrtr@r num_framesheightwidthoutput_num_frames output_height output_widthcount output_latentvae_batch_inputrjkn_startn_endh_starth_endw_startw_end video_cubelatent output_kernel output_strideoutput_overlap latent_means r5r(!AutoencoderKLAllegro.tiled_encodes  + + [[ 3 3>?gg; *e JJ [[^(Ct{{ST~'UVYZZ F[[^$;t{{1~#MNQRR zz5;;q>#9T[[^"KLqP  !1L@DKK/// A"$ A"$ A"$   ++'7t{{ST~W[WbWbcdWegkgrgrstgu&vw()A=)|,A%&Q%7T[[^9KdkkZ[n9\UG%&Q%7T[[^9KdkkZ[n9\UG%&Q%7T[[^9KdkkZ[n9\UG!"1a wu}#T!UJ@JO,<$<=04Dq4HH $E $TWX$XX!%o!>"%F%UXY%YY % 8!+<=L.*;<%a-&7,&F\IY&Y\]&]^hhijk #K 1a wu}LMQ\\M-**1aA.66q!<(!!!j"%56>>q!Q1M r7c  SnURnURRnURSU-URSU-URSU-4nURSU-URSU-URSU-4nUR upxpn UR SSSSS5RSS5nURU5nURSUS45R SSSSS5n[R"XS- US- 5S-n [R"XS- US- 5S-n [R"XS- US- 5S-nSnURX-U-URRURSURSURS45nURX(USUSUS45n[U 5GH n[U 5Hn[U5HnUUS-UUS-US-nnUUS-UUS-US-nnUUS-UUS-US-nnUSS2SS2UU2UU2UU24nUUX-'X-US- :XdXU -U-S- :XahURU5nXU -U-S- :XaX-US- :waUSX-S-UXU-- S&O UUX- S-US-&URX(USUSUS45nUS- nM M GM URXpRRX-X-X-45nURSURS- URSURS- URSURS- 4n[U 5GH!nUURS-UURS-URS-nn[U 5HnUURS-UURS-URS-nn[U5HnUURS-UURS-URS-nn[!UU US4UU US4UUUS4UUU -U-UU--U-R#S55nUSS2SS2UU2UU2UU24==U- ss'M M GM$ UR SSSSS5R%5nU$rD)rrErrrrHrFrIr rErFrGrHr_rwr rIrJ contiguous) r0r7rKrLrM latent_kernel latent_strider@rrNrOrPrQrRrSrTdecoded_videosrVrrWrXrYrZr[r\r]r^current_latent current_videovideo video_overlapout_video_blends r5r9!AutoencoderKLAllegro.tiled_decodes  + + [[ 3 3 A",dkk!n.BDKKPQNVXDXX  A",dkk!n.BDKKPQNVXDXX >?gg; *e IIaAq! $ , ,Q 2   # KKJ+ , 4 4Q1a C JJ 15E(EWXIY'YZ]^^ F1-=$=qAQ#QRUVV zz5+;#;}Q?O"OPSTT !1L@ (( A A A   ++ ]1-=}Q?OQ^_`Qa b ()A=)|,A%&q)9%91}Q?O;OR_`aRb;bUG%&q)9%91}Q?O;OR_`aRb;bUG%&q)9%91}Q?O;OR_`aRb;bUG%&q!WU]GEM7SX='X%YN@NOE$<=04Dq4HH $E $TWX$XX(, _(E "%F%UXY%YY % 8%":Q">RN5;K3K+K+MNXeN5+Ca+G%RS)T*+++-]1=M}]^O_anopaqr+QJE9-**@ Z)A)A:?TZT_afaklm KKNT[[^ + KKNT[[^ + KKNT[[^ + ()AQ/T[[^1CdkkRSn1TUG=)!"T[[^!3QQ5G$++VW.5X|,A%&Q%7T[[^9KdkkZ[n9\UG&8-}Q/?@M=+;<L-*:;&q='8<'G!lJZ'Z]^'^_iijkl 'O !Q wu}gemKLP__L-** aAq!,779 r7rsample_posterior generatorcUnURU5RnU(aURUS9nOUR5nUR U5RnU(dU4$[ US9$)a Args: sample (`torch.Tensor`): Input sample. sample_posterior (`bool`, *optional*, defaults to `False`): Whether to sample from the posterior. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`DecoderOutput`] instead of a plain tuple. generator (`torch.Generator`, *optional*): PyTorch random number generator. )rrr=)r5r/rmoderAr ) r0rrqr-rrr&r4r7decs r5rKAutoencoderKLAllegro.forward.sf" KKN..   9 5A Akk!n##6MC((r7) r r rr r rrrrrrr)rrrrrrrrBr rrNrBi@gp= ף?T)rN)T)FTN)#rOrPrQrRrS _supports_gradient_checkpointingrr%rrrUrTr$rrr!r$r<rWr*r rrrr5r:r rAr(r9r GeneratorrKrXrYrZs@r5rrs+Z(,$- + /C7Q5O !!,- $!5G G G  S/ G c3hG "#s(O!G "%*$)$4#G $#(c "2%G &'G ()G *+G ,-G .%*/G 01G 23G 45G 6 7G G R  ![[%,,[37::,0: "E*F$GG H::8[[%,,[- -4-5X]XdXdIdCe--2QellQu||QfWellWu||Wx"' /3 ) )) ) EOO, ) }ell* + ))r7rcUupEnUupxn Uupn US:GaUS:ayUSS2SS2SU2SS2SS24[R"SU5R5RUR5U- R USS5-USS2SS2SU2SS2SS24'XES- :a~USS2SS2U*S2SS2SS24S[R"SU5R5RUR5U- - R USS5-USS2SS2U*S2SS2SS24'US:axUSS2SS2SS2SU 2SS24[R"SU 5R5RUR5U - R U S5-USS2SS2SS2SU 2SS24'XxS- :a}USS2SS2SS2U *S2SS24S[R"SU 5R5RUR5U - - R U S5-USS2SS2SS2U *S2SS24'U S:ahUSS2SS2SS2SS2SU 24[R"SU 5R5RUR5U - -USS2SS2SS2SS2SU 24'XS- :amUSS2SS2SS2SS2U *S24S[R"SU 5R5RUR5U - - -USS2SS2SS2SS2U *S24'U$)Nrr )r<arangerTrdevicereshape) n_paramh_paramw_paramr&nn_max overlap_nr3h_max overlap_hww_max overlap_ws r5rIrIMs!Ai!Ai!Ai1} q5)*1a9a+B)C Q *00255ahh?)KgiA&*'AaAiKA% & qy=)*1a)a+B)CELLI.44699!((CiOOgiA&*'AaYJKA% & 1u%&q!Q) Q'>%? LLI & , , . 1 1!(( ;i G ')Q & !Q1Y; !" 19}%&q!Q Q'>%?  Q *00255ahh?)K K ')Q & !QI:; !" 1u%&q!Q1Y;'>%? LLI & , , . 1 1!(( ;i G& !Q1a k !" 19}%&q!QI:;'>%?  Q *00255ahh?)K K& !Q1yjk !" Hr7)&rFtypingrrrr<torch.nnr&configuration_utilsrrutils.accelerate_utilsr attention_processorr r autoencoders.vaer r downsamplingrmodeling_outputsrmodeling_utilsrrr upsamplingrModulerr\rrrrrrIr7r5rs  )) B88J'2'"#bryybJTTnOryyOdlBIIl^|ryy|~DryyDNf):{f)R  r7