biddlmZmZmZmZmZmZddlZddlmZddl m Z m Z ddl m Z ddlmZdd lmZmZmZmZmZdd lmZmZmZmZdd lmZdd lmZmZdd l m!Z!ddl"m#Z#m$Z$m%Z%ejLe'Z(GddejRZ*GddejRZ+GddejRZ,Gddee e Z-y))AnyDictListOptionalTupleUnionN)nn) ConfigMixinregister_to_config)FromOriginalModelMixin)logging)ADDED_KV_ATTENTION_PROCESSORSCROSS_ATTENTION_PROCESSORSAttentionProcessorAttnAddedKVProcessor AttnProcessor)TextImageTimeEmbeddingTextTimeEmbeddingTimestepEmbedding Timesteps) ModelMixin)UNetMidBlock2DCrossAttnget_down_block)UNet2DConditionModel)ControlNetConditioningEmbeddingControlNetOutput zero_modulecHeZdZdZdej dej fdZy) QuickGELUzr Applies GELU approximation that is fast but somewhat inaccurate. See: https://github.com/hendrycks/GELUs inputreturnc8|tjd|zzS)NgZd;?)torchsigmoid)selfr#s h/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/models/controlnets/controlnet_union.pyforwardzQuickGELU.forward/su}}UU]333N)__name__ __module__ __qualname____doc__r&Tensorr*r+r)r"r"*s"4U\\4ell4r+r"cDeZdZdeffd Zdej fdZxZS)ResidualAttentionMlpd_modelct|tj||dz|_t |_tj|dz||_y)N)super__init__r Linearc_fcr"geluc_proj)r(r4 __class__s r)r8zResidualAttentionMlp.__init__4sC IIgw{3 K ii! W5 r+xcl|j|}|j|}|j|}|SN)r:r;r<r(r>s r)r*zResidualAttentionMlp.forward:s. IIaL IIaL KKNr+) r,r-r.intr8r&r0r* __classcell__r=s@r)r3r33s66 r+r3ceZdZddededej ffd Zdej fdZdej fdZxZ S) ResidualAttentionBlockr4n_head attn_maskct|tj|||_tj ||_t||_tj ||_ ||_ yr@) r7r8r MultiheadAttentionattn LayerNormln_1r3mlpln_2rH)r(r4rGrHr=s r)r8zResidualAttentionBlock.__init__BsU ))'6: LL) '0LL) "r+r>c|j1|jj|j|jnd|_|j |||d|jdS)NdtypedeviceF) need_weightsrHr)rHtorRrSrKrAs r) attentionz ResidualAttentionBlock.attentionJsTNRnnNh***JnryyAquyOPQRRr+c||j|j|z}||j|j|z}|Sr@)rVrMrNrOrAs r)r*zResidualAttentionBlock.forwardNs= tyy|, , 1& &r+r@) r,r-r.rBr&r0r8rVr*rCrDs@r)rFrFAsC##S#U\\#S5<<Sr+rFcJeZdZdZdZe dDdededededee d fd e eeefd eed fd ed ede de de ede dede eeed ffde ede e de eeed ffde e eeed ffdede e de e de ede edede de ed e d!e eed fd"ed#ed$ed%ed&ed'ed(efHfd) Z e dEd*ed e d!e eed fd+efd,Zed-ee effd.Zd/e eee efffd0Zd1Zd2e e eeefd-dfd3Z dFd4ej4d5e ej4e efd6ej4d7eej4d8ej4d9eed:e e ee fd;e ej4de ee ej4fd?e ee efd@edAedBed-e eeeej4d fej4fff dCZxZS)GControlNetUnionModela A ControlNetUnion model. Args: in_channels (`int`, defaults to 4): The number of channels in the input sample. flip_sin_to_cos (`bool`, defaults to `True`): Whether to flip the sin to cos in the time embedding. freq_shift (`int`, defaults to 0): The frequency shift to apply to the time embedding. down_block_types (`tuple[str]`, defaults to `("CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D")`): The tuple of downsample blocks to use. only_cross_attention (`Union[bool, Tuple[bool]]`, defaults to `False`): block_out_channels (`tuple[int]`, defaults to `(320, 640, 1280, 1280)`): The tuple of output channels for each block. layers_per_block (`int`, defaults to 2): The number of layers per block. downsample_padding (`int`, defaults to 1): The padding to use for the downsampling convolution. mid_block_scale_factor (`float`, defaults to 1): The scale factor to use for the mid block. act_fn (`str`, defaults to "silu"): The activation function to use. norm_num_groups (`int`, *optional*, defaults to 32): The number of groups to use for the normalization. If None, normalization and activation layers is skipped in post-processing. norm_eps (`float`, defaults to 1e-5): The epsilon to use for the normalization. cross_attention_dim (`int`, defaults to 1280): The dimension of the cross attention features. transformer_layers_per_block (`int` or `Tuple[int]`, *optional*, defaults to 1): The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`]. Only relevant for [`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`], [`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`]. encoder_hid_dim (`int`, *optional*, defaults to None): If `encoder_hid_dim_type` is defined, `encoder_hidden_states` will be projected from `encoder_hid_dim` dimension to `cross_attention_dim`. encoder_hid_dim_type (`str`, *optional*, defaults to `None`): If given, the `encoder_hidden_states` and potentially other embeddings are down-projected to text embeddings of dimension `cross_attention` according to `encoder_hid_dim_type`. attention_head_dim (`Union[int, Tuple[int]]`, defaults to 8): The dimension of the attention heads. use_linear_projection (`bool`, defaults to `False`): class_embed_type (`str`, *optional*, defaults to `None`): The type of class embedding to use which is ultimately summed with the time embeddings. Choose from None, `"timestep"`, `"identity"`, `"projection"`, or `"simple_projection"`. addition_embed_type (`str`, *optional*, defaults to `None`): Configures an optional embedding which will be summed with the time embeddings. Choose from `None` or "text". "text" will use the `TextTimeEmbedding` layer. num_class_embeds (`int`, *optional*, defaults to 0): Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing class conditioning with `class_embed_type` equal to `None`. upcast_attention (`bool`, defaults to `False`): resnet_time_scale_shift (`str`, defaults to `"default"`): Time scale shift config for ResNet blocks (see `ResnetBlock2D`). Choose from `default` or `scale_shift`. projection_class_embeddings_input_dim (`int`, *optional*, defaults to `None`): The dimension of the `class_labels` input when `class_embed_type="projection"`. Required when `class_embed_type="projection"`. controlnet_conditioning_channel_order (`str`, defaults to `"rgb"`): The channel order of conditional image. Will convert to `rgb` if it's `bgr`. conditioning_embedding_out_channels (`tuple[int]`, *optional*, defaults to `(48, 96, 192, 384)`): The tuple of output channel for each block in the `conditioning_embedding` layer. global_pool_conditions (`bool`, defaults to `False`): TN in_channelsconditioning_channelsflip_sin_to_cos freq_shiftdown_block_types.only_cross_attentionblock_out_channelslayers_per_blockdownsample_paddingmid_block_scale_factoract_fnnorm_num_groupsnorm_epscross_attention_dimtransformer_layers_per_blockencoder_hid_dimencoder_hid_dim_typeattention_head_dimnum_attention_headsuse_linear_projectionclass_embed_typeaddition_embed_typeaddition_time_embed_dimnum_class_embedsupcast_attentionresnet_time_scale_shift%projection_class_embeddings_input_dim%controlnet_conditioning_channel_order#conditioning_embedding_out_channelsglobal_pool_conditionsaddition_embed_type_num_headsnum_control_typenum_trans_channelnum_trans_headnum_trans_layernum_proj_channelc% t4||xs|}t|t|k7rtd|d|dt |t s)t|t|k7rtd|d|dt |t s)t|t|k7rtd|d|dt |t r|gt|z}d}%|%dz dz}&tj||d |%|& |_ |d d z}'t|d |||_ |d }(t|(|'| |_ |td |dd|_||tj||'|_nc|dk(rt|(|'|_nL|dk(rtj"|'|'|_n+|dk(r| tdt||'|_nd|_|dk(r||})n|})t%|)|'||_nS|dk(rt)|||'|_n:|dk(r$t||||_t||'|_n|td|dt-|d |||_|!dz}*tj0|*t3j4| |!z|_tj8t;|#D+cgc]}+t=|!|"c}+|_tAtjB|!|$|_"t||||_#t|| z|'|_$tj8g|_%tj8g|_&t |t r|gt|z}t |t r|ft|z}t |t r|ft|z}|d },tj|,|,d}-tA|-}-|jLjO|-tQ|D]\}.}/|,}0||.},|.t|dz k(}1tS|/fid|d||.d|0d |,d!|'d"|1 d#| d$| d%| d&|d'||.d(||.||.n|,d)| d*|d+||.d,|d-|}2|jJjO|2t;|D]@}+tj|,|,d}-tA|-}-|jLjO|-B|1rtj|,|,d}-tA|-}-|jLjO|-|d.}3tj|3|3d}-tA|-}-|-|_*tW|d.|3|'| | | |||d.| ||/ |_,ycc}+w)0NzbMust provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: z. `down_block_types`: .zfMust provide the same number of `only_cross_attention` as `down_block_types`. `only_cross_attention`: zdMust provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: r rrr) kernel_sizepaddingr6)rdzencoder_hid_dim_type: z must be None.timestepidentity projectionzX`class_embed_type`: 'projection' requires `projection_class_embeddings_input_dim` be settext) num_heads text_image)text_embed_dimimage_embed_dimtime_embed_dim text_timezaddition_embed_type: z& must be None, 'text' or 'text_image'.)conditioning_embedding_channelsr`r[g?)r num_layersrhrZ out_channels temb_channelsadd_downsample resnet_eps resnet_act_fn resnet_groupsrgrlrkrbrmr_rrrs) rhrZrrroutput_scale_factorrsrgrlrrmrr)-r7r8len ValueError isinstanceboolrBr Conv2dconv_inr time_projrtime_embeddingencoder_hid_proj Embeddingclass_embeddingIdentityr add_embeddingr add_time_projrcontrolnet_cond_embedding Parameterr&randntask_embedding ModuleListrangerFtransformer_layesr r9spatial_ch_projscontrol_type_projcontrol_add_embedding down_blockscontrolnet_down_blocksappend enumeratercontrolnet_mid_blockr mid_block)5r(rZr[r\r]r^r_r`rarbrcrdrerfrgrhrirjrkrlrmrnrorprqrrrsrtrurvrwrxryrzr{r|r}conv_in_kernelconv_in_paddingrtimestep_input_dimtext_time_embedding_from_dimtask_scale_factor_output_channelcontrolnet_blockidown_block_type input_channelis_final_block down_blockmid_block_channelr=s5 r)r8zControlNetUnionModel.__init__sZ 2G5G ! "c*:&; ;tvHuII_`p_qqrs .5#>R:SWZ[kWl:lxzNyOOefvewwxy -s3Z[ [$(D !  #(8(D#%<<0@.#QD  +#45G#XD  +#%;;~~#ND  -4< n$55Z\j#kD #'D & (*/>,/B,!2,nHe"D !L 0"82DWhv"D !K /!*+BOU_!`D !23XZh!iD  ,45H4IIopq q*I,>q,AB"7* & .s2 ll+j9|j>j;| j@r3| j@j9|j@j;| jBj9|jBj;d | jDj9|jDj;d | S)"a3 Instantiate a [`ControlNetUnionModel`] from [`UNet2DConditionModel`]. Parameters: unet (`UNet2DConditionModel`): The UNet model weights to copy to the [`ControlNetUnionModel`]. All configuration options are also copied where applicable. rhrriNrjrorprZr\r]r^r_r`rarbrcrdrerfrgrkrlrmrnrqrrrsrtrurvF)strictr1)#configrhrirjrorprZr\r]r^r_r`rarbrcrdrerfrgrkrlrmrnrqrrrsrtrload_state_dict state_dictrrrrr) clsrrurvrrhrirjrorp controlnets r) from_unetzControlNetUnionModel.from_unets"9WZ^ZeZe8eDKK 4 4kl %:Kdkk9Y$++55_cCY]a]h]hCht{{??nrAVZ^ZeZeAedkk==ko3LPTP[P[3[DKK / /ae  + !5 !4 %<  *F   //  !KK77 {{-- "[[99 "&!A!A  ${{== "[[99  ${{== $(;;#E#E ;;%% !KK77! "[[))# $!% ? ?% & ${{==' (!% ? ?) *#'++"C"C+ ,"[[99- ."[[99/ 0"[[991 2%)KK$G$G3 437++2c2c5 63X7 81T9 > "    . .t||/F/F/H I  0 01J1J1L M  % % 5 5d6I6I6T6T6V W))**::4;O;O;Z;Z;\]  " " 2 243C3C3N3N3PY^ 2 _  0 01J1J1LUZ 0 [r+r$ci}dtdtjjdttt fffd |j D]\}}||||S)z Returns: `dict` of attention processors: A dictionary containing all attention processors used in the model with indexed by its weight name. namemodule processorsct|dr|j||d<|jD]\}}|d||||S)N get_processor .processorr)hasattrrnamed_children)rrrsub_namechildfn_recursive_add_processorss r)rzIControlNetUnionModel.attn_processors..fn_recursive_add_processorssdv/282F2F2H dV:./#)#8#8#: U%+tfAhZ,@%T U r+)strr&r Modulerrr)r(rrrrs @r)attn_processorsz$ControlNetUnionModel.attn_processorssm  c 588?? X\]`bt]tXu !//1 BLD& 'fj A Br+ processorc Tt|jj}t|tr,t||k7rt dt|d|d|ddt dtjjffd |jD]\}}|||y) a4 Sets the attention processor to use to compute attention. Parameters: processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`): The instantiated processor class or a dictionary of processor classes that will be set as the processor for **all** `Attention` layers. If `processor` is a dict, the key needs to define the path to the corresponding cross attention processor. This is strongly recommended when setting trainable attention processors. z>A dict of processors was passed, but the number of processors z0 does not match the number of attention layers: z. Please make sure to pass z processor classes.rrct|drEt|ts|j|n#|j|j |d|j D]\}}|d|||y)N set_processorrr)rrdictrpopr)rrrrrfn_recursive_attn_processors r)rzLControlNetUnionModel.set_attn_processor..fn_recursive_attn_processorsxv/!)T2((3(($z7J)KL#)#8#8#: T%+tfAhZ,@%S Tr+N) rrkeysrrrrr&r rr)r(rcountrrrs @r)set_attn_processorz'ControlNetUnionModel.set_attn_processorsD((--/0 i &3y>U+BPQTU^Q_P`a005w6QRWQXXkm   Tc T588?? T!//1 ALD& 'fi @ Ar+c jtd|jjDr t}nmtd|jjDr t }n8t dt t|jj|j|y)ze Disables custom attention processors and sets the default attention implementation. c3@K|]}|jtvywr@)r=r.0procs r) zBControlNetUnionModel.set_default_attn_processor.. si4t~~!>>ic3@K|]}|jtvywr@)r=rrs r)rzBControlNetUnionModel.set_default_attn_processor.. sh$#==hrzOCannot call `set_default_attn_processor` when attention processors are of type N) allrvaluesrrrnextiterr)r(rs r)set_default_attn_processorz/ControlNetUnionModel.set_default_attn_processors i4K_K_KfKfKhi i,.I h$J^J^JeJeJgh h%IabfgklpmAmAmHmHmJhKcLbMN   *r+ slice_sizec  g dtjjf fd |jD] }| t }|dk(r Dcgc]}|dz }}n |dk(r|dgz}t |t s||gzn|}t |t k7r=tdt |d|jd t d t d tt |D]&}||} |}|||kDstd |d|d dtjjdttf fd t t|}|jD] } || y cc}w)a Enable sliced attention computation. When this option is enabled, the attention module splits the input tensor in slices to compute attention in several steps. This is useful for saving some memory in exchange for a small decrease in speed. Args: slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`): When `"auto"`, input to the attention heads is halved, so attention is computed in two steps. If `"max"`, maximum amount of memory is saved by running only one slice at a time. If a number is provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim` must be a multiple of `slice_size`. rct|drj|j|jD] }| yNset_attention_slice)rrsliceable_head_dimchildren)rr$fn_recursive_retrieve_sliceable_dimssliceable_head_dimss r)rzVControlNetUnionModel.set_attention_slice..fn_recursive_retrieve_sliceable_dims's@v45#**6+D+DE* <4U; .fn_recursive_set_attention_sliceMsEv45**:>>+;<* D0 C Dr+) r&r rrrrlistrrrrrBreversed) r(rrnum_sliceable_layersdimrsizereversed_slice_sizerrrs @@@r)rz(ControlNetUnionModel.set_attention_slices! < <mmo 9F 0 8 9 ##67  /BBs#(BJB 5 -3J@J:W[@\)ZL8bl z?c"56 6$S_$5VDKK=cReNfMghQQTUhQiPjjkm  s:' VAa=D%a(CD3J 5.McURS!TUU  V DUXX__ DRVWZR[ D#8J#78mmo JF ,V5H I J=Cs E;samplerencoder_hidden_statescontrolnet_cond control_typecontrol_type_idxconditioning_scale class_labels timestep_condattention_maskadded_cond_kwargscross_attention_kwargs from_multi guess_mode return_dictcxt|tr|gt|z}|jj}|dk7rt d|| 2d| j |jz dz} | jd} |}tj|s|jjdk(}|jjdk(}t|tr%|s|rtjntj}n$|s|rtjntj }tj"|g||j}n6t|j$d k(r|dj |j}|j'|j$d }|j)|}|j |j }|j+|| }d}|j,g| t d |jj.d k(r|j)|}|j-|j |j }||z}|jj0"|jj0d k(r|j3|}n|jj0dk(rd| vrt |j4d| j7d}d| vrt |j4d| j7d}|j9|j;}|j=|j$d df}tj>||gd}|j |j}|j3|}|jA|j;}|j=|j$d df}|j |j}|jC|}||z}|||zn|}|jE|}g}g} tG|||D]\}!}"}#|jI|!}$tjJ|$d}%|%|jL|"z}%| st|dk(r2|jO|%jd| jO|$|jO|%jd|#z| jO|$|#z|}$tjJ|$d}%|jO|%jd| jO|$tjP|d}&|jRD] }'|'|&}& |dz}(tGtU| dd|D]d\\})}$}#|jW|&dd|)f}*|*jdjd}*| st|dk(r |(|$|*zz }(Z|(|$|*|#zzz }(f||(z}|f}+|jXD]=},t[|,dr|,j\r|,|||| | \}}-n |,||\}}-|+|-z }+?|j^|j_|||| | }d}.tG|+|j`D]\}/}0|0|/}/|.|/fz}.|.}+|jc|}1|r|jjdsqtjfdd t|+dz|j}2| st|dk(r|2|d z}2tG|+|2D#cgc] \}}#||#z }+}}#|1|2dz}1n-| st|dk(r|+Dcgc] }||d z }+}|1|d z}1|jjdr=|+Dcgc]}tjJ|dd}+}tjJ|1dd}1|s|+|1fSti|+|1Scc}#}wcc}wcc}w) am The [`ControlNetUnionModel`] forward method. Args: sample (`torch.Tensor`): The noisy input tensor. timestep (`Union[torch.Tensor, float, int]`): The number of timesteps to denoise an input. encoder_hidden_states (`torch.Tensor`): The encoder hidden states. controlnet_cond (`List[torch.Tensor]`): The conditional input tensors. control_type (`torch.Tensor`): A tensor of shape `(batch, num_control_type)` with values `0` or `1` depending on whether the control type is used. control_type_idx (`List[int]`): The indices of `control_type`. conditioning_scale (`float`, defaults to `1.0`): The scale factor for ControlNet outputs. class_labels (`torch.Tensor`, *optional*, defaults to `None`): Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings. timestep_cond (`torch.Tensor`, *optional*, defaults to `None`): Additional conditional embeddings for timestep. If provided, the embeddings will be summed with the timestep_embedding passed through the `self.time_embedding` layer to obtain the final timestep embeddings. attention_mask (`torch.Tensor`, *optional*, defaults to `None`): An attention mask of shape `(batch, key_tokens)` is applied to `encoder_hidden_states`. If `1` the mask is kept, otherwise if `0` it is discarded. Mask will be converted into a bias, which adds large negative values to the attention scores corresponding to "discard" tokens. added_cond_kwargs (`dict`): Additional conditions for the Stable Diffusion XL UNet. cross_attention_kwargs (`dict[str]`, *optional*, defaults to `None`): A kwargs dictionary that if specified is passed along to the `AttnProcessor`. from_multi (`bool`, defaults to `False`): Use standard scaling when called from `MultiControlNetUnionModel`. guess_mode (`bool`, defaults to `False`): In this mode, the ControlNet encoder tries its best to recognize the input content of the input even if you remove all prompts. A `guidance_scale` between 3.0 and 5.0 is recommended. return_dict (`bool`, defaults to `True`): Whether or not to return a [`~models.controlnet.ControlNetOutput`] instead of a plain tuple. Returns: [`~models.controlnet.ControlNetOutput`] **or** `tuple`: If `return_dict` is `True`, a [`~models.controlnet.ControlNetOutput`] is returned, otherwise a tuple is returned where the first element is the sample tensor. rgbz1unknown `controlnet_conditioning_channel_order`: NrgmpsnpurQr)rRz9class_labels should be provided when num_class_embeds > 0rrr text_embedsz has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`time_idsz has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`r)r)rr ghas_cross_attention) hidden_statestembrr r)rr)rr rr1)rST)rkeepdim)down_block_res_samplesmid_block_res_sample)5rfloatrrrurrUrR unsqueezer& is_tensorrStypefloat32float64int32int64tensorshapeexpandrrrrnrorr=getrflattenreshapeconcatrrrziprmeanrrcatrrrrrrrrrrwlogspacer)3r(rrrrrr r r r r rrrrr channel_order timestepsis_mpsis_npurRt_embembaug_emb class_embrr time_embeds add_embedscontrol_embeds control_embinputscondition_listcond control_idxscale conditionfeat_seqr>layercontrolnet_cond_fuseridxalphardownsample_block res_samples!controlnet_down_block_res_samplesdown_block_res_samplerrscaless3 r)r*zControlNetUnionModel.forwardXs@ (% 0"4!5O8L!L  II E !PQ^P_`a a  %."3"3FLL"AAXMN+55a8N y)]]''50F]]''50F(E**0F (.& u{{ i[fmmTI  !Q &!$**6==9I$$V\\!_5 y) v||,!!%7    +# !\]]{{++z9#~~l; ,,\:==DJJ=OI /C ;; * * 6{{..&8,,-BC00K? (99$>>*+@033MB %66$>>*+|}-00<"001A1A1CD )11;3D3DQ3G2LM "\\; *D"M ']]3995 ,,Z8// 0D0D0FG'//Q0DE'**399500@ K&2cGmf%(+O=MOa(b 9 $D+u66t?DTDhDh&6"(*?#1+A '# '7VRU&V# "k 1 " 2 >> %^^&;-'= $F-/)7:;QSWSnSn7o m 3 !#3$45J$K !0QUjTl0l - m"C#88@ dkk@@^^B3/E+F+JSYS`S`aFS!12a7"4Q"77JMNdflJm%nfun%n "%n#7&*#D 3/0A5Si%jf/A!/D&D%j "%j#7:LQ:O#O ;; - -Ka&AG 6vt<& "&$)::.BX\#] *,@A A#9Pd  &o&k&s>^,-^2 ^7)$r6r Tr)CrossAttnDownBlock2DrNrN DownBlock2DF)@irQrrrsilu gh㈵>rQrNNNFNNNNFdefaultNr)0`iF@rPrTrrP)r)rSrWT) g?NNNNNFFT)r,r-r.r/ _supports_gradient_checkpointingr rBrrrrrrr8 classmethodrrpropertyrrrrrrrr&r0rrr*rCrDs@r)rYrYTs?B(,$%& $- :?.D !"#())+#'DE)-.2:;EI&+*.-115*.!&'0?C5:I[',-/ !!$ #Ui i  #i  i  i  S/ i $D%+$56i "#s(Oi i  i !&!i "#i $"#%i &'i (!)i *',CsCx,@&A+i ,"#-i .'sm/i 0"#uS#X"671i 2&eCsCx,@&AB3i 4 $5i 6#3-7i 8&c]9i :"*#;i <#3-=i >?i @"%Ai B08}Ci D03Ei F.6eCHo-FGi H!%Ii J(+Ki LMi NOi PQi RSi TUi i V6;IZ'+ C"C03C.6eCHo-F C !% CCJc+=&=!>0 AE2Dd3PbKbFc2c,d AF + ?JeCd3i4G.H?JT?JR9'?'M!NN O#q r+rY).typingrrrrrrr&r configuration_utilsr r loaders.single_file_modelr utilsrattention_processorrrrrr embeddingsrrrrmodeling_utilsrunets.unet_2d_blocksrrunets.unet_2d_conditionrrrrr get_loggerr,loggerrr"r3rFrYr1r+r)rks;: B?a`';VV   H %4 4 299 RYY&u  :{4Ju  r+