oMjS SSKrSSKrSSKJr SSKJrJr SSKJr SSK J r SSK J r SSK Jr SS KJr SS KJrJrJrJrJrJr SS KJr SS KJr S SKJrJrJrJ r J!r! "SS\\\\\ \5r"g)N) ConfigMixinregister_to_config)PeftAdapterMixin)FromOriginalModelMixin) deprecate)apply_forward_hook)AttentionMixin)ADDED_KV_ATTENTION_PROCESSORSCROSS_ATTENTION_PROCESSORS AttentionAttnAddedKVProcessor AttnProcessorFusedAttnProcessor2_0)AutoencoderKLOutput) ModelMixin)AutoencoderMixinDecoder DecoderOutputDiagonalGaussianDistributionEncoderc(^\rSrSrSrSrSS/r\S3S\S\S \ \ S 4S \ \ S 4S \ \S 4S \ \S 4S-S\S\ S\S\S\S\ S\ S\ S\ S\ S\ S\ \\44$U4Sjjj5r SrS\R S\R 4Sjr\S4S\R S\ S\\ \-4S jj5rS4S!\R S\ S\\R -4S"jjr\S5S!\R0S\ S\\R0-4S#jj5rS$\R S%\R S&\S\R 4S'jrS$\R S%\R S&\S\R 4S(jrS\R S\R 4S)jrS4S\R S\ S\4S*jjrS4S!\R S\ S\\R -4S+jjrS6S,\R S-\ S\ S.\R>S-S\\R -4 S/jjr S0r!S1r"S2r#U=r$$)7AutoencoderKLFlux2&a A VAE model with KL loss for encoding images into latents and decoding latent representations into images. 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, *optional*, defaults to 3): Number of channels in the input image. out_channels (int, *optional*, defaults to 3): Number of channels in the output. down_block_types (`tuple[str]`, *optional*, defaults to `("DownEncoderBlock2D",)`): Tuple of downsample block types. up_block_types (`tuple[str]`, *optional*, defaults to `("UpDecoderBlock2D",)`): Tuple of upsample block types. block_out_channels (`tuple[int]`, *optional*, defaults to `(64,)`): Tuple of block output channels. act_fn (`str`, *optional*, defaults to `"silu"`): The activation function to use. latent_channels (`int`, *optional*, defaults to 4): Number of channels in the latent space. sample_size (`int`, *optional*, defaults to `32`): Sample input size. force_upcast (`bool`, *optional*, 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 mid_block_add_attention (`bool`, *optional*, default to `True`): If enabled, the mid_block of the Encoder and Decoder will have attention blocks. If set to false, the mid_block will only have resnet blocks TBasicTransformerBlock ResnetBlock2DN in_channels out_channelsdown_block_types.up_block_typesblock_out_channelsdecoder_block_out_channelslayers_per_blockact_fnlatent_channelsnorm_num_groups sample_size force_upcastuse_quant_convuse_post_quant_convmid_block_add_attentionbatch_norm_epsbatch_norm_momentum patch_sizec >[TU]5 [UU UUUUU SUS9 Ul[ U UUU=(d UUU UUS9UlU (a[ R"SU -SU -S5OSUlU(a[ R"XS5OSUl [ R"[R"U5U -UUSSS9Ul SUlSUlUR R"Ul['UR R"[([*45(aUR R"SOUR R"n [-U S[/UR R05S- -- 5UlS Ulg) NT) rr r!r#r%r&r(double_zr-)rr r"r#r%r(r&r-r rF)epsmomentumaffinetrack_running_statsrg?)super__init__rencoderrdecodernnConv2d quant_convpost_quant_conv BatchNorm2dmathprodbn use_slicing use_tilingconfigr)tile_sample_min_size isinstancelisttupleintlenr#tile_latent_min_sizetile_overlap_factor)selfrr r!r"r#r$r%r&r'r(r)r*r+r,r-r.r/r0 __class__s C:\Users\johnn\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages\diffusers/models/autoencoders/autoencoder_kl_flux2.pyr8AutoencoderKLFlux2.__init__GspJ #(-1-+$;   '%)9O=O-+$;   Uc"))A$7_9LaPhlQdryy1Mjn.. IIj !O 3( $  !%)KK$;$;!$++11D%=AA KK # #A &((  %( qSA_A_=`cd=d7e(f$g!#' c ~[SURR555(a [5nOr[SURR555(a [ 5nO8[ S[ [URR55535eURU5 g)zU Disables custom attention processors and sets the default attention implementation. c3F# UHoR[;v M g7fN)rOr .0procs rP @AutoencoderKLFlux2.set_default_attn_processor..siKh4~~!>>Kh!c3F# UHoR[;v M g7frU)rOr rVs rPrYrZshJg$#==Jgr[zOCannot call `set_default_attn_processor` when attention processors are of type N) allattn_processorsvaluesrr ValueErrornextiterset_attn_processor)rN processors rPset_default_attn_processor-AutoencoderKLFlux2.set_default_attn_processors i4K_K_KfKfKhi i i,.I h$J^J^JeJeJgh h h%IabfgklpmAmAmHmHmJhKcLbMN   *rRxreturncURup#pEUR(a/XPR:dX@R:aURU5$UR U5nUR bUR U5nU$rU)shaperDrF _tiled_encoder9r=)rNrg batch_size num_channelsheightwidthencs rP_encodeAutoencoderKLFlux2._encodesh23''/ & ??(A(A AVNgNgEg%%a( (ll1o ?? &//#&C rR return_dictc>UR(aYURSS:aFURS5Vs/sHo0RU5PM nn[R "U5nOURU5n[ U5nU(dU4$[US9$s snf)a Encode a batch of images into latents. Args: x (`torch.Tensor`): Input batch of images. return_dict (`bool`, *optional*, defaults to `True`): Whether to return a [`~models.autoencoder_kl.AutoencoderKLOutput`] instead of a plain tuple. Returns: The latent representations of the encoded images. If `return_dict` is True, a [`~models.autoencoder_kl.AutoencoderKLOutput`] is returned, otherwise a plain `tuple` is returned. rr latent_dist)rCrjsplitrqtorchcatrr)rNrgrsx_sliceencoded_slicesh posteriors rPencodeAutoencoderKLFlux2.encodes    QCD771:N:ll73:NN .)A QA03 < "y99OsBzc:UR(aIURSUR:dURSUR:aURXS9$URbUR U5nUR U5nU(dU4$[ US9$)N)rssample)rDrjrL tiled_decoder>r:r)rNrrsdecs rP_decodeAutoencoderKLFlux2._decodes ?? d.G.G G177SU;Y]YrYrKr$$Q$@ @    +$$Q'All1o6MC((rRcPUR(acURSS:aPURS5Vs/sHo@RU5RPM nn[ R "U5nOURU5RnU(dU4$[US9$s snf)a Decode a batch of images. Args: z (`torch.Tensor`): Input batch of latent vectors. return_dict (`bool`, *optional*, 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. rrr)rCrjrwrrrxryr)rNrrs generatorz_slicedecoded_slicesdecodeds rPdecodeAutoencoderKLFlux2.decodes$    QJK''RS*U*wll73::*NUii/Gll1o,,G: G,,Vs$B#ab blend_extentc[URSURSU5n[U5H@nUSS2SS2U*U-SS24SXC- - -USS2SS2USS24XC- --USS2SS2USS24'MB U$)Nr rminrjrange)rNrrrys rPblend_vAutoencoderKLFlux2.blend_vs1771:qwwqz<@ |$Aa\MA$5q89QAQ=QRUVWXZ[]^`aWaUbfgfvUwwAaAqjM%rRc[URSURSU5n[U5H@nUSS2SS2SS2U*U-4SXC- - -USS2SS2SS2U4XC- --USS2SS2SS2U4'MB U$)Nrrr)rNrrrrgs rPblend_hAutoencoderKLFlux2.blend_h s1771:qwwqz<@ |$AaA }q'889QAQ=QRUVWXZ[]^`aWaUbfgfvUwwAaAqjM%rRc [URSUR- -5n[URUR-5nURU- n/n[ SUR SU5Hn/n[ SUR SU5HznUSS2SS2XfUR-2XUR-24n UR U 5n URR(aURU 5n URU 5 M| URU5 M /n [U5Hupg/n [U5HbupUS:aURXVS- UX5n US:aURXxS- X5n U RU SS2SS2SU2SU245 Md U R[R"U SS95 M [R"U SS9n U $)aEncode a batch of images using a tiled encoder. When this option is enabled, the VAE will split the input tensor into tiles to compute encoding in several steps. This is useful to keep memory use constant regardless of image size. The end result of tiled encoding is different from non-tiled encoding because each tile uses a different encoder. To avoid tiling artifacts, the tiles overlap and are blended together to form a smooth output. You may still see tile-sized changes in the output, but they should be much less noticeable. Args: x (`torch.Tensor`): Input batch of images. Returns: `torch.Tensor`: The latent representation of the encoded videos. rrr rNdim)rJrFrMrLrrjr9rEr+r=append enumeraterrrxry) rNrg overlap_sizer row_limitrowsirowjtile result_rows result_rowrps rPrk AutoencoderKLFlux2._tiled_encodes"444Dq5<<U AKDq5<<E DGD!!$q!ZiZ)'C"DE*   uyy; <&ii + rRc &Sn[SSUSS9 [URSUR- -5n[URUR-5nURU- n/n[ SUR SU5Hn/n [ SUR S U5Hzn US S 2S S 2XUR-2XUR-24n URU 5n URR(aURU 5n U RU 5 M| URU 5 M /n [U5Hup/n [U 5HbupUS:aURXxS- U X5n U S:aURXS- X5n U RU S S 2S S 2S U2S U245 Md U R[R "U S S 95 M [R "U SS 9n[#U5nU(dU4$[%US 9$) aEncode a batch of images using a tiled encoder. When this option is enabled, the VAE will split the input tensor into tiles to compute encoding in several steps. This is useful to keep memory use constant regardless of image size. The end result of tiled encoding is different from non-tiled encoding because each tile uses a different encoder. To avoid tiling artifacts, the tiles overlap and are blended together to form a smooth output. You may still see tile-sized changes in the output, but they should be much less noticeable. Args: x (`torch.Tensor`): Input batch of images. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~models.autoencoder_kl.AutoencoderKLOutput`] instead of a plain tuple. Returns: [`~models.autoencoder_kl.AutoencoderKLOutput`] or `tuple`: If return_dict is True, a [`~models.autoencoder_kl.AutoencoderKLOutput`] is returned, otherwise a plain `tuple` is returned. aCThe tiled_encode implementation supporting the `return_dict` parameter is deprecated. In the future, the implementation of this method will be replaced with that of `_tiled_encode` and you will no longer be able to pass `return_dict`. You will also have to create a `DiagonalGaussianDistribution()` from the returned value. tiled_encodez1.0.0F) standard_warnrrr rNrru)rrJrFrMrLrrjr9rEr+r=rrrrrxryrr)rNrgrsdeprecation_messagerrrrrrrrrrmomentsr}s rPrAutoencoderKLFlux2.tiled_encode?s( ~  .'+>eT444Dq5<<U AKDq5<<E DGD!!$q!ZiZ)'C"DE*   uyy; <&))KQ/09 < "y99rRc [URSUR- -5n[URUR-5nURU- n/n[ SUR SU5Hn/n[ SUR SU5Hzn USS2SS2XwUR-2XUR-24n UR R(aURU 5n URU 5n URU 5 M| URU5 M /n [U5Hupx/n [U5HbupUS:aURXgS- U X5n U S:aURXS- X5n U RU SS2SS2SU2SU245 Md U R[R"U SS95 M [R"U SS9nU(dU4$[!US9$)a Decode a batch of images using a tiled decoder. Args: z (`torch.Tensor`): Input batch of latent vectors. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~models.vae.DecoderOutput`] instead of a plain tuple. Returns: [`~models.vae.DecoderOutput`] or `tuple`: If return_dict is True, a [`~models.vae.DecoderOutput`] is returned, otherwise a plain `tuple` is returned. rrr rNrr)rJrLrMrFrrjrEr,r>r:rrrrrxryr)rNrrsrrrrrrrrrrrrs rPrAutoencoderKLFlux2.tiled_decode}s444Dq5<<U AKDq5<<E DGD!!$q!ZiZ)'C"DE*   uyy; <&ii +6MC((rRrsample_posteriorrcUnURU5RnU(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*): A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make sampling deterministic. 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~rvrmoderr) rNrrrsrrgr}rrs rPforwardAutoencoderKLFlux2.forwardsf. KKN..   9 5A Akk!n##6MC((rRcSUlURR5H3upS[URR 5;dM*[ S5e URUlUR5H)n[U[5(dMURSS9 M+ UR[55 g)u Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value) are fused. For cross-attention modules, key and value projection matrices are fused. > [!WARNING] > This API is 🧪 experimental. NAddedzQ`fuse_qkv_projections()` is not supported for models having added KV projections.T)fuse) original_attn_processorsr^itemsstrrO__name__r`modulesrGrfuse_projectionsrcr)rN_attn_processormodules rPfuse_qkv_projections'AutoencoderKLFlux2.fuse_qkv_projectionss)-%!%!5!5!;!;!= A#n66??@@ !tuu">)-(<(<%llnF&),,''T'2%  5 78rRcVURbURUR5 gg)u^Disables the fused QKV projection if enabled. > [!WARNING] > This API is 🧪 experimental. N)rrc)rNs rPunfuse_qkv_projections)AutoencoderKLFlux2.unfuse_qkv_projectionss)  ( ( 4  # #D$A$A B 5rR) rBr:r9rr>r=rLrMrFrCrD)rr)DownEncoderBlock2Drrr)UpDecoderBlock2Drrr)rNr silu riTTTTg-C6?g?)r r )T)TN)FTN)%r __module__ __qualname____firstlineno____doc__ _supports_gradient_checkpointing_no_split_modulesrrJrIrboolfloatr8rerxTensorrqr rrr~rr FloatTensorrrrrkrr Generatorrrr__static_attributes__ __classcell__)rOs@rPrr&sZ6(,$0/B- + / >B !!!!#$((, $%(&,EU(U(U( S/ U(c3hU( "#s(O!U(,%*#s(Od$:-U(./U(01U(23U(45U(67U(89U(:;U(<"=U(>"&?U(@AU(B#CU(D#s(OEU(U(p +  %,, 37::,0: u%AB B::8 ) )D )MTYT`T`D` )HL-""-15- ** *--8%,,cell %,,cell .u||. .`<:ell<:<:I\<:|/)ell/)/)Y^YeYeIe/)h"' ,0 ") ")") ") ??T) ")  % ")J9,CCrRr)#r@rxtorch.nnr;configuration_utilsrrloadersrloaders.single_file_modelrutilsrutils.accelerate_utilsr attentionr attention_processorr r rrrrmodeling_outputsrmodeling_utilsrvaerrrrrrrRrPrsX B'?8&3'``JC .+?UWgJCrR