3j,SSKJr SSKrSSKJr SSKJs Jr SSKJ r SSK J r SSK J r SSKJrJrJrJrJr SS KJr SS KJrJrJrJrJrJr "S S \R:5r"S S\R:5rS\R@S\R@4Sjr!"SS\R:5r""SS\R:5r#"SS\R:5r$"SS\R:5r%"SS\R:5r&"SS\R:5r'g))partialN) deprecate)get_activation) SpatialNorm) Downsample1D Downsample2DFirDownsample2D KDownsample2D downsample_2d) AdaGroupNorm) FirUpsample2D KUpsample2D Upsample1D Upsample2Dupfirdn2d_native upsample_2dc"^\rSrSrSrSSSSSSSS S S SSSS SS .S\S\S-S\S\S\S\S\S-S\S\S\S\S\S-S\S\S\S\S-4 U4Sjjjr S\ RS \ RS!\ R4S"jr S#r U=r$)$ResnetBlockCondNorm2D+a A Resnet block that use normalization layer that incorporate conditioning information. Parameters: in_channels (`int`): The number of channels in the input. out_channels (`int`, *optional*, default to be `None`): The number of output channels for the first conv2d layer. If None, same as `in_channels`. dropout (`float`, *optional*, defaults to `0.0`): The dropout probability to use. temb_channels (`int`, *optional*, default to `512`): the number of channels in timestep embedding. groups (`int`, *optional*, default to `32`): The number of groups to use for the first normalization layer. groups_out (`int`, *optional*, default to None): The number of groups to use for the second normalization layer. if set to None, same as `groups`. eps (`float`, *optional*, defaults to `1e-6`): The epsilon to use for the normalization. non_linearity (`str`, *optional*, default to `"swish"`): the activation function to use. time_embedding_norm (`str`, *optional*, default to `"ada_group"` ): The normalization layer for time embedding `temb`. Currently only support "ada_group" or "spatial". kernel (`torch.Tensor`, optional, default to None): FIR filter, see [`~models.resnet.FirUpsample2D`] and [`~models.resnet.FirDownsample2D`]. output_scale_factor (`float`, *optional*, default to be `1.0`): the scale factor to use for the output. use_in_shortcut (`bool`, *optional*, default to `True`): If `True`, add a 1x1 nn.conv2d layer for skip-connection. up (`bool`, *optional*, default to `False`): If `True`, add an upsample layer. down (`bool`, *optional*, default to `False`): If `True`, add a downsample layer. conv_shortcut_bias (`bool`, *optional*, default to `True`): If `True`, adds a learnable bias to the `conv_shortcut` output. conv_2d_out_channels (`int`, *optional*, default to `None`): the number of channels in the output. If None, same as `out_channels`. NF ư>swish ada_group?T) out_channels conv_shortcutdropout temb_channelsgroups groups_outeps non_linearitytime_embedding_normoutput_scale_factoruse_in_shortcutupdownconv_shortcut_biasconv_2d_out_channels in_channelsrr r!r"r#r$r%r&r'r(r)r*r+r,r-c $>[TU]5 XlUcUOUnX lX0lXlXlXlXlUcUnURS:Xa[XQXhS9Ul O9URS:Xa[X5Ul O[SUR35e[R"XSSSS9UlURS:Xa[XRXxS9UlO9URS:Xa[X%5UlO[SUR35e["RR%U5UlU=(d Un[R"UUSSSS9Ul[+U 5UlS=UlUlUR (a[3USS 9UlO"UR (a[5USSS S 9UlU cURU:gOU UlSUlUR6(a[R"UUSSS US 9Ulgg)Nr)r%spatialz" unsupported time_embedding_norm: r kernel_sizestridepaddingFuse_convopr7r5namerr3r4r5bias)super__init__r.ruse_conv_shortcutr*r+r(r'rnorm1r ValueErrornnConv2dconv1norm2torchDropoutr!conv2r nonlinearityupsample downsamplerr r)r )selfr.rr r!r"r#r$r%r&r'r(r)r*r+r,r- __class__s Q/home/wildlama/miniconda3/lib/python3.13/site-packages/diffusers/models/resnet.pyr>ResnetBlockCondNorm2D.__init__Is( &&2&:{ (!. #6 #6  J  # #{ 2%m&RDJ  % % 2$[@DJA$BZBZA[\] ]YY{aPQ[\]  # #{ 2%m:WDJ  % % 2$\ADJA$BZBZA[\] ]xx''0 3C|YY|-AqYZdef *=9*..  77&{UCDM YY*;PQX\]DOKZKbt//3GGhw!   !#$' "D  input_tensortembreturnc[U5S:dURSS5bSn[SSU5 UnURXb5nUR U5nUR bVUR SS:a UR5nUR5nUR U5nUR U5nO/URb"URU5nURU5nURU5nURXb5nUR U5nURU5nURU5nURbURU5nX-UR- nU$)NrscaleThe `scale` argument is deprecated and will be ignored. Please remove it, as passing it will raise an error in the future. `scale` should directly be passed while calling the underlying pipeline component i.e., via `cross_attention_kwargs`.1.0.0@)lengetrr@rIrJshape contiguousrKrDrEr!rHr r()rLrQrRargskwargsdeprecation_message hidden_states output_tensors rNforwardResnetBlockCondNorm2D.forwardsN t9q=FJJw5A#U  gw(; <$  =7 ))-8 == $""1%++668 - 8 8 : ==6L MM-8M __ (??<8L OOM:M =1  =7 ))-8  ]3  =1    )--l;L%59Q9QQ rP)rDrHr r+rKr!r.rIr@rErr(r'r*rJr?r))__name__ __module__ __qualname____firstlineno____doc__intboolfloatstrr>rFTensorrb__static_attributes__ __classcell__rMs@rNrr+s3B$(# !%$#.%('+#'+/%IIDj I  I  III$JIII!I#II I !I"!#I$"Dj%IIV%ELL% %Z_ZfZf%%rPrc(*^\rSrSrSrSSSSSSSS S SS SS SSSSSS .S\S\S-S\S\S\S\S\S-S\S\S\S\S\S\ RS-S\S\S-S\S\S\S \S-4&U4S!jjjr S"\ RS#\ RS$\ R4S%jr S&r U=r$)' ResnetBlock2Da A Resnet block. Parameters: in_channels (`int`): The number of channels in the input. out_channels (`int`, *optional*, default to be `None`): The number of output channels for the first conv2d layer. If None, same as `in_channels`. dropout (`float`, *optional*, defaults to `0.0`): The dropout probability to use. temb_channels (`int`, *optional*, default to `512`): the number of channels in timestep embedding. groups (`int`, *optional*, default to `32`): The number of groups to use for the first normalization layer. groups_out (`int`, *optional*, default to None): The number of groups to use for the second normalization layer. if set to None, same as `groups`. eps (`float`, *optional*, defaults to `1e-6`): The epsilon to use for the normalization. non_linearity (`str`, *optional*, default to `"swish"`): the activation function to use. time_embedding_norm (`str`, *optional*, default to `"default"` ): Time scale shift config. By default, apply timestep embedding conditioning with a simple shift mechanism. Choose "scale_shift" for a stronger conditioning with scale and shift. kernel (`torch.Tensor`, optional, default to None): FIR filter, see [`~models.resnet.FirUpsample2D`] and [`~models.resnet.FirDownsample2D`]. output_scale_factor (`float`, *optional*, default to be `1.0`): the scale factor to use for the output. use_in_shortcut (`bool`, *optional*, default to `True`): If `True`, add a 1x1 nn.conv2d layer for skip-connection. up (`bool`, *optional*, default to `False`): If `True`, add an upsample layer. down (`bool`, *optional*, default to `False`): If `True`, add a downsample layer. conv_shortcut_bias (`bool`, *optional*, default to `True`): If `True`, adds a learnable bias to the `conv_shortcut` output. conv_2d_out_channels (`int`, *optional*, default to `None`): the number of channels in the output. If None, same as `out_channels`. NFrrrTrrdefaultr)rr r!r"r#r$pre_normr%r& skip_time_actr'kernelr(r)r*r+r,r-r.rr r!r"r#r$rur%r&rvr'rwr(r)r*r+r,r-c >^[TU]5 U S:Xa [S5eU S:Xa [S5eSUlXlUcUOUnX lX0lUUlUUlXl Xl Xl UcUn[RRXaU SS9Ul[R "XSSSS 9UlUbuURS :Xa[R$"XR5UlOPURS :Xa [R$"US U-5UlO [S URS35eSUl[RRXrU SS9Ul[RR+U5UlU=(d Un[R "UUSSSS 9Ul[1U 5UlS=UlUlUR(aIU S:XaSmU4SjUlOU S:Xa[9[:R<SSS9UlOk[?USS9UlO[UR(aJU S:XaSmU4SjUlO6U S:Xa[9[:R@S S S9UlO[CUSSSS9UlUcURU:gOUUl"SUl#URD(a[R "UUSSSUS9Ul#gg)NrzkThis class cannot be used with `time_embedding_norm==ada_group`, please use `ResnetBlockCondNorm2D` insteadr0ziThis class cannot be used with `time_embedding_norm==spatial`, please use `ResnetBlockCondNorm2D` insteadT num_groups num_channelsr%affiner1rr2rt scale_shiftrzunknown time_embedding_norm :  fir)rr1r1rc>[UTS9$N)rw)rx fir_kernels rN(ResnetBlock2D.__init__..$s +a *KrPsde_vpg@nearest) scale_factormodeFr6c>[UTS9$r)r rs rNrr,s M!J,OrP)r3r4r8r9rr;)$r=r>rArur.rr?r*r+r(r'rvrFrB GroupNormr@rCrDLinear time_emb_projrErGr!rHrrIrJrKrF interpolater avg_pool2dr r)r )rLr.rr r!r"r#r$rur%r&rvr'rwr(r)r*r+r,r-rrMs @rNr>ResnetBlock2D.__init__s.  + -}  ) +{  &&2&:{ (!. #6 #6 *  JXX''6Y\ei'j YY{aPQ[\]  $''94%'YY}%K"))]:%'YY}a,>N%O" #A$BZBZA[[\!]^^!%D XX'':^ajn'o xx''0 3C|YY|-AqYZdef *=9*..  77) K 8# ' Ci X *; G YY) "O8#")!,,Aa"P".{UTU\`"aKZKbt//3GGhw!   !#$' "D  rPrQrRrSc[U5S:dURSS5bSn[SSU5 UnURU5nUR U5nUR bVUR SS:a UR5nUR5nUR U5nUR U5nO/URb"URU5nURU5nURU5nURb>UR(dUR U5nURU5SS2SS2SS4nURS:XaUbXb-nURU5nOqURS:XaPUc[SUR35e[R "US S S 9upxURU5nUS U--U-nOURU5nUR U5nUR#U5nUR%U5nUR&b2UR((aUR5nUR'U5nX-UR*- n U $) NrrUrVrWrXrtr}z9 `temb` should not be None when `time_embedding_norm` is rr)dim)rYrZrr@rIrJr[r\rKrDrrvr'rErArFchunkr!rHr trainingr() rLrQrRr]r^r_r` time_scale time_shiftras rNrbResnetBlock2D.forward?s< t9q=FJJw5A#U  gw(; <$  =1 ))-8 == $""1%++668 - 8 8 : ==6L MM-8M __ (??<8L OOM:M =1    )%%((.%%d+Aq$,<=D  # #y 0 - 4  JJ}5M  % % 6| OPTPhPhOij&+[[qa%@ "J JJ}5M)Q^rbrnrorps@rNrrrrsfD$(# !%$##,&*%('+#'+/+bbDj b  b  bbb$Jbbbbb!b t#b #!b"#b$ %b&'b(!)b*"Dj+bbH:ELL: :Z_ZfZf::rPrrtensorrSc[UR5S:Xa USS2SS2S4$[UR5S:XaUSS2SS2SSS24$[UR5S:XaUSS2SS2SSS24$[S[U5S35e)Nrr1rz`len(tensor)`: z has to be 2, 3 or 4.)rYr[rA)rs rNrearrange_dimsr}s 6<<AaDj!! 6<<AaD!m$$ V\\ a aAqj!!?3v;-7LMNNrPc ^\rSrSrSrS S\S\S\\\\4-S\S\4 U4SjjjrS \ RS \ R4S jr S r U=r $) Conv1dBlockiax Conv1d --> GroupNorm --> Mish Parameters: inp_channels (`int`): Number of input channels. out_channels (`int`): Number of output channels. kernel_size (`int` or `tuple`): Size of the convolving kernel. n_groups (`int`, default `8`): Number of groups to separate the channels into. activation (`str`, defaults to `mish`): Name of the activation function. inp_channelsrr3n_groups activationc>[TU]5 [R"XX3S-S9Ul[R "XB5Ul[U5Ulg)Nrr5) r=r>rBConv1dconv1dr group_normrmish)rLrrr3rrrMs rNr>Conv1dBlock.__init__sD ii K`aQab ,,x>":. rPinputsrScURU5n[U5nURU5n[U5nURU5nU$N)rrrr)rLrintermediate_reproutputs rNrbConv1dBlock.forwardsM KK/*+<= OO,=>*+<=,- rP)rrr)rrdrerfrgrhrituplerlr>rFrmrbrnrorps@rNrrsw   / / /5c?* /  /  / /ellu||rPrc ^\rSrSrSrSS\S\S\S\\\\4-S\4 U4SjjjrS \ RS \ RS \ R4S jr S r U=r $)ResidualTemporalBlock1Diau Residual 1D block with temporal convolutions. Parameters: inp_channels (`int`): Number of input channels. out_channels (`int`): Number of output channels. embed_dim (`int`): Embedding dimension. kernel_size (`int` or `tuple`): Size of the convolving kernel. activation (`str`, defaults `mish`): It is possible to choose the right activation function. rr embed_dimr3rc4>[TU]5 [XU5Ul[X"U5Ul[ U5Ul[R"X25Ul X:wa[R"XS5Ul g[R"5Ul g)Nr) r=r>rconv_inconv_outr time_emb_actrBrtime_embrIdentity residual_conv)rLrrrr3rrMs rNr> ResidualTemporalBlock1D.__init__s{ "<{K #L L *:6 ): 9E8TBIIl! 4 Z\ZeZeZg rPrtrScURU5nURU5nURU5[U5-nUR U5nX0R U5-$)z Args: inputs : [ batch_size x inp_channels x horizon ] t : [ batch_size x embed_dim ] returns: out : [ batch_size x out_channels x horizon ] )rrrrrr)rLrrouts rNrbResidualTemporalBlock1D.forwards\   a  MM! ll6"^A%66mmC ''///rP)rrrrr)rrrps@rNrrs  ./      5c?*     & 0ell 0u|| 0 0 0rPrc ^\rSrSrSrSS\S\S-S\S\4U4SjjjrSS \RS \S \R4S jjr S r U=r $)TemporalConvLayeria Temporal convolutional layer that can be used for video (sequence of images) input Code mostly copied from: https://github.com/modelscope/modelscope/blob/1509fdb973e5871f37148a4b5e5964cafd43e64d/modelscope/models/multi_modal/video_synthesis/unet_sd.py#L1016 Parameters: in_dim (`int`): Number of input channels. out_dim (`int`): Number of output channels. dropout (`float`, *optional*, defaults to `0.0`): The dropout probability to use. Nin_dimout_dimr!norm_num_groupsc f>[TU]5 U=(d UnXlX l[R "[R "XA5[R"5[R"XSSS95Ul [R "[R "XB5[R"5[R"U5[R"X!SSS95Ul [R "[R "XB5[R"5[R"U5[R"X!SSS95Ul [R "[R "XB5[R"5[R"U5[R"X!SSS95Ul [RRURSR 5 [RRURSR"5 g)Nr1rr)rrrr)r=r>rrrB SequentialrSiLUConv3drDrGrHconv3conv4initzeros_weightr<)rLrrr!rrMs rNr>TemporalConvLayer.__init__sb #V  ]] LL 1 GGI IIfy) D  ]] LL 2 GGI JJw  IIgy) D   ]] LL 2 GGI JJw  IIgy) D   ]] LL 2 GGI JJw  IIgy) D    tzz"~,,- tzz"~**+rPr` num_framesrScUSSS24RSU4URSS-5RSSSSS5nUnURU5nUR U5nUR U5nUR U5nX1-nURSSSSS5RURSURS-S4URSS-5nU$)Nrrrrr1r)reshaper[permuterDrHrr)rLr`ridentitys rNrbTemporalConvLayer.forwards $' " * *B +;m>Q>QRSRT>U+U V ^ ^_`bcefhikl m ! =1  =1  =1  =1  0 %--aAq!<DD   #m&9&9!&< rFrmrbrnrorps@rNrrso#! ',',t', ',  ',',RU\\s5<<rPrc ^\rSrSrSrSS\S\S-S\S\4U4SjjjrS \RS \RS \R4S jr S r U=r $)TemporalResnetBlocki"a A Resnet block. Parameters: in_channels (`int`): The number of channels in the input. out_channels (`int`, *optional*, default to be `None`): The number of output channels for the first conv2d layer. If None, same as `in_channels`. temb_channels (`int`, *optional*, default to `512`): the number of channels in timestep embedding. eps (`float`, *optional*, defaults to `1e-6`): The epsilon to use for the normalization. Nr.rr"r%c>[TU]5 XlUcUOUnX lSnUVs/sHofS-PM nn[R R SXSS9Ul[ R"UUUSUS9Ul Ub[ R"X25Ul OSUl [R R SX$SS9Ul [R RS5Ul[ R"UUUSUS9Ul[!S 5UlURU:gUlSUlUR$(a[ R"UUSSS S9Ulggs snf) NrrrTryrr2rsilur)r=r>r.rrFrBrr@rrDrrrErGr!rHrrIr)r ) rLr.rr"r%r3kr5rMs rNr>TemporalResnetBlock.__init__.sR &&2&:{ ( #./;a6;/XX''2Kae'f YY  #     $!#=!GD !%D XX''2Lbf'g xx'', YY  #   +62#//<?!   !# "D  A0sE rQrRrScUnURU5nURU5nURU5nURbIURU5nURU5SS2SS2SS2SS4nUR SSSSS5nX2-nUR U5nURU5nUR U5nURU5nURbURU5nX-nU$)Nrrrr1r) r@rIrDrrrEr!rHr )rLrQrRr`ras rNrbTemporalResnetBlock.forwardds$  =1 ))-8  =1    )$$T*D%%d+Aq!T4,?@D<<1aA.D)0M =1 ))-8  ]3  =1    )--l;L$4 rP) rDrHr r!r.rIr@rErrr))Nrrrrps@rNrr"ss $(  44Dj4 4  44lELL rPrc^\rSrSrSrSS\S\S-S\S\S\S-S \S \4U4S jjjrSS \ RS \ RS-S\ RS-4Sjjr Sr U=r $)SpatioTemporalResBlockia A SpatioTemporal Resnet block. Parameters: in_channels (`int`): The number of channels in the input. out_channels (`int`, *optional*, default to be `None`): The number of output channels for the first conv2d layer. If None, same as `in_channels`. temb_channels (`int`, *optional*, default to `512`): the number of channels in timestep embedding. eps (`float`, *optional*, defaults to `1e-6`): The epsilon to use for the spatial resenet. temporal_eps (`float`, *optional*, defaults to `eps`): The epsilon to use for the temporal resnet. merge_factor (`float`, *optional*, defaults to `0.5`): The merge factor to use for the temporal mixing. merge_strategy (`str`, *optional*, defaults to `learned_with_images`): The merge strategy to use for the temporal mixing. switch_spatial_to_temporal_mix (`bool`, *optional*, defaults to `False`): If `True`, switch the spatial and temporal mixing. Nr.rr"r% temporal_eps merge_factorswitch_spatial_to_temporal_mixc >[T U]5 [UUUUS9Ul[ UbUOUUbUOUUUbUOUS9Ul[ UUUS9Ulg)N)r.rr"r%)alphamerge_strategyr)r=r>rrspatial_res_blockrtemporal_res_block AlphaBlender time_mixer) rLr.rr"r%rrrrrMs rNr>SpatioTemporalResBlock.__init__sp !.#%' " #6(4(@ k)5)A{' , 8 c # ')+I rPr`rRimage_only_indicatorcURSnURX5nURupVpxXT-n USSS24RXXgU5RSSSSS5n USSS24RXXgU5RSSSSS5nUbURXS5nUR X5nUR U UUS9nURSSSSS5RXVXx5nU$)Nrrrrr1r) x_spatial x_temporalr)r[rrrrr) rLr`rRrr batch_frameschannelsheightwidth batch_sizehidden_states_mixs rNrbSpatioTemporalResBlock.forwards$ *//3 ..}C 0=0C0C- !/  $' " * *:8UZ [ c cdeghjkmnpq r  $' " * *:8UZ [ c cdeghjkmnpq r   << ;D// D '$!5( &--aAq!<DD\]ck rP)rrr)NrrNg?learned_with_imagesF)NN)rdrerfrgrhrirkrjr>rFrmrbrnrorps@rNrrs($( %)!,/4  Dj     dl   )-  H%)48 ||llT!$llT1 rPrc ^\rSrSrSr/SQrSS\S\S\4U4Sjjjr S\ RS \ S \ R4S jr SS \ RS\ RS\ RS -S \ R4SjjrSrU=r$)ria A module to blend spatial and temporal features. Parameters: alpha (`float`): The initial value of the blending factor. merge_strategy (`str`, *optional*, defaults to `learned_with_images`): The merge strategy to use for the temporal mixing. switch_spatial_to_temporal_mix (`bool`, *optional*, defaults to `False`): If `True`, switch the spatial and temporal mixing. )learnedfixedrrrrc>[TU]5 X lX0lX R;a[ SUR35eURS:Xa(UR S[R"U/55 gURS:XdURS:XaEURS[RR[R"U/555 g[ SUR35e)Nzmerge_strategy needs to be in r mix_factorrrzUnknown merge strategy ) r=r>rr strategiesrAregister_bufferrFrmregister_parameterrB Parameter)rLrrrrMs rNr>AlphaBlender.__init__s ,.L+  0=doo=NOP P   ' )  u||UG/D E  I -1D1DH]1]  # #L%((2D2DU\\SXRYEZ2[ \6t7J7J6KLM MrPrndimsrSc @URS:XaURnU$URS:Xa"[R"UR5nU$URS:XaUc [ S5e[R "UR 5[R"SSURS9[R"UR5S5nUS:XaUSS2SSS2SS4nU$US :XaURS 5SS2SS4nU$[ S US 35e[e) NrrrzMPlease provide image_only_indicator to use learned_with_images merge strategyr)device).Nrr1rzUnexpected ndims z. Dimensions should be 3 or 5) rrrFsigmoidrAwhererjonesrrNotImplementedError)rLrrrs rN get_alphaAlphaBlender.get_alphas    ' )OOE6 3 I -MM$//2E0 - $9 9#+ !pqqKK$))+ 1a(<(C(CD doo.y9Ezaq$45 ! b)!T4-8 !#4UG;X!YZZ& %rPNrrcURX1R5nURUR5nUR(aSU- nXA-SU- U--nU$)Nr)r ndimtodtyper)rLrrrrrs rNrbAlphaBlender.forwardsV 3^^D)  . .%KE  u : :rP)rr)rFr)rdrerfrgrhrrkrlrjr>rFrmrir rbrnrorps@rNrrs =J 4/4 NNN)- NN(ell35<<F59 << LL $llT1    rPr)( functoolsrrFtorch.nnrBtorch.nn.functional functionalrutilsr activationsrattention_processorr downsamplingr r r r r normalizationr upsamplingrrrrrrModulerrrrmrrrrrrrrPrNr s  ',(NBIINb}BII}BO5<<OELLO ")) H,0bii,0^D DNY"))YzQRYYQhN299NrP