biddlZddlmZmZmZmZddlZddlZddl m Z m Z ddl m Z mZddlmZddlmZmZmZerddlZ d d ZGd d ee Zy) N)ListOptionalTupleUnion) ConfigMixinregister_to_config) deprecateis_scipy_available) randn_tensor)KarrasDiffusionSchedulersSchedulerMixinSchedulerOutputc $|dk(rd}n|dk(rd}ntd|g}t|D]<}||z }|dz|z }|jtd||||z z |>t j |tj S)a Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [0,1]. Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up to that part of the diffusion process. Args: num_diffusion_timesteps (`int`): the number of betas to produce. max_beta (`float`): the maximum beta to use; use values lower than 1 to prevent singularities. alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar. Choose from `cosine` or `exp` Returns: betas (`np.ndarray`): the betas used by the scheduler to step the model outputs cosinecftj|dzdz tjzdz dzS)NgMb?gT㥛 ?r)mathcospits v/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/schedulers/scheduling_dpmsolver_multistep_inverse.py alpha_bar_fnz)betas_for_alpha_bar..alpha_bar_fn;s-88QY%/$''9A=>!C Cexpc2tj|dzS)Ng()rrrs rrz)betas_for_alpha_bar..alpha_bar_fn@s88AI& &rz"Unsupported alpha_transform_type: r dtype) ValueErrorrangeappendmintorchtensorfloat32)num_diffusion_timestepsmax_betaalpha_transform_typerbetasit1t2s rbetas_for_alpha_barr."s.x' D  & '=>R=STUU E * +M ( (!e. . S\"- R0@@@(KLM <<U]] 33rc0eZdZdZeDcgc]}|j c}}ZdZeddddddd d d d d ddd d d d d d e d dddfde de de de de e ejee fde de dede de de de ded ed!e ed"e ed#e ed$e ed%e e d&e d'e e d(e d)e f.d*Zed+ZdMd,e d-e e ej,ffd.Zd/ej0d0ej0fd1Zd2Zd3Zd4ej0d0ej0fd5Zd4ej0d,e d0ej0fd6Z dNd4ej0d,e d7e d8e d0ej0f d9Zdd:d;ej0d/ej0d0ej0fd<Zddd=d;ej0d/ej0d>e ej0d0ej0fd?Z ddd=d@eej0d/ej0d>e ej0d0ej0fdAZ!ddd=d@eej0d/ej0d>e ej0d0ej0fdBZ"dCZ# dOd;ej0dDe e ej0fd/ej0dEe ej0dFed0e e$e%ff dGZ&d/ej0d0ej0fdHZ'dIej0d>ej0dJejPd0ej0fdKZ)dLZ*ycc}}w)P"DPMSolverMultistepInverseScheduleru `DPMSolverMultistepInverseScheduler` is the reverse scheduler of [`DPMSolverMultistepScheduler`]. This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving. Args: num_train_timesteps (`int`, defaults to 1000): The number of diffusion steps to train the model. beta_start (`float`, defaults to 0.0001): The starting `beta` value of inference. beta_end (`float`, defaults to 0.02): The final `beta` value. beta_schedule (`str`, defaults to `"linear"`): The beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from `linear`, `scaled_linear`, or `squaredcos_cap_v2`. trained_betas (`np.ndarray`, *optional*): Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`. solver_order (`int`, defaults to 2): The DPMSolver order which can be `1` or `2` or `3`. It is recommended to use `solver_order=2` for guided sampling, and `solver_order=3` for unconditional sampling. prediction_type (`str`, defaults to `epsilon`, *optional*): Prediction type of the scheduler function; can be `epsilon` (predicts the noise of the diffusion process), `sample` (directly predicts the noisy sample`) or `v_prediction` (see section 2.4 of [Imagen Video](https://imagen.research.google/video/paper.pdf) paper). thresholding (`bool`, defaults to `False`): Whether to use the "dynamic thresholding" method. This is unsuitable for latent-space diffusion models such as Stable Diffusion. dynamic_thresholding_ratio (`float`, defaults to 0.995): The ratio for the dynamic thresholding method. Valid only when `thresholding=True`. sample_max_value (`float`, defaults to 1.0): The threshold value for dynamic thresholding. Valid only when `thresholding=True` and `algorithm_type="dpmsolver++"`. algorithm_type (`str`, defaults to `dpmsolver++`): Algorithm type for the solver; can be `dpmsolver`, `dpmsolver++`, `sde-dpmsolver` or `sde-dpmsolver++`. The `dpmsolver` type implements the algorithms in the [DPMSolver](https://huggingface.co/papers/2206.00927) paper, and the `dpmsolver++` type implements the algorithms in the [DPMSolver++](https://huggingface.co/papers/2211.01095) paper. It is recommended to use `dpmsolver++` or `sde-dpmsolver++` with `solver_order=2` for guided sampling like in Stable Diffusion. solver_type (`str`, defaults to `midpoint`): Solver type for the second-order solver; can be `midpoint` or `heun`. The solver type slightly affects the sample quality, especially for a small number of steps. It is recommended to use `midpoint` solvers. lower_order_final (`bool`, defaults to `True`): Whether to use lower-order solvers in the final steps. Only valid for < 15 inference steps. This can stabilize the sampling of DPMSolver for steps < 15, especially for steps <= 10. euler_at_final (`bool`, defaults to `False`): Whether to use Euler's method in the final step. It is a trade-off between numerical stability and detail richness. This can stabilize the sampling of the SDE variant of DPMSolver for small number of inference steps, but sometimes may result in blurring. use_karras_sigmas (`bool`, *optional*, defaults to `False`): Whether to use Karras sigmas for step sizes in the noise schedule during the sampling process. If `True`, the sigmas are determined according to a sequence of noise levels {σi}. use_exponential_sigmas (`bool`, *optional*, defaults to `False`): Whether to use exponential sigmas for step sizes in the noise schedule during the sampling process. use_beta_sigmas (`bool`, *optional*, defaults to `False`): Whether to use beta sigmas for step sizes in the noise schedule during the sampling process. Refer to [Beta Sampling is All You Need](https://huggingface.co/papers/2407.12173) for more information. lambda_min_clipped (`float`, defaults to `-inf`): Clipping threshold for the minimum value of `lambda(t)` for numerical stability. This is critical for the cosine (`squaredcos_cap_v2`) noise schedule. variance_type (`str`, *optional*): Set to "learned" or "learned_range" for diffusion models that predict variance. If set, the model's output contains the predicted Gaussian variance. timestep_spacing (`str`, defaults to `"linspace"`): The way the timesteps should be scaled. Refer to Table 2 of the [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://huggingface.co/papers/2305.08891) for more information. steps_offset (`int`, defaults to 0): An offset added to the inference steps, as required by some model families. r ig-C6?g{Gz?linearNrepsilonFgףp= ?? dpmsolver++midpointTinflinspacernum_train_timesteps beta_startbeta_end beta_schedule trained_betas solver_orderprediction_type thresholdingdynamic_thresholding_ratiosample_max_valuealgorithm_type solver_typelower_order_finaleuler_at_finaluse_karras_sigmasuse_exponential_sigmasuse_beta_sigmasuse_flow_sigmas flow_shiftlambda_min_clipped variance_typetimestep_spacing steps_offsetc|jjrts tdt |jj|jj |jj gdkDr td| dvrd| d}tdd||+tj|tj |_ n|d k(r-tj|||tj |_ nk|d k(r6tj|d z|d z|tj d z|_ n0|dk(rt||_ nt|d|j d|jz |_tj$|j"d|_tj(|j&|_tj(d|j&z |_tj.|j*tj.|j,z |_d|j&z |j&z d z|_d|_| dvr2| dk(r|j7dnt| d|j | dvr1| dvr|j7dnt| d|j d|_t;jd|dz |t:j j=}tj>||_ dg|z|_!d|_"d|_#|j2jId|_||_||_||_y)Nz:Make sure to install scipy if you want to use beta sigmas.r znOnly one of `config.use_beta_sigmas`, `config.use_exponential_sigmas`, `config.use_karras_sigmas` can be used. dpmsolver sde-dpmsolverzalgorithm_type zn is deprecated and will be removed in a future version. Choose from `dpmsolver++` or `sde-dpmsolver++` insteadz+algorithm_types dpmsolver and sde-dpmsolver1.0.0rr1 scaled_linear?rsquaredcos_cap_v2z is not implemented for r3rdim)rQr4rRsde-dpmsolver++deisr4)rB)r5heun)logrhobh1bh2r5)rCcpu)%configrHr ImportErrorsumrGrFr r r$r%r&r*r7r.NotImplementedError __class__alphascumprodalphas_cumprodsqrtalpha_tsigma_tloglambda_tsigmasinit_noise_sigmar num_inference_stepsnpcopy from_numpy timesteps model_outputslower_order_nums _step_indexto)selfr8r9r:r;r<r=r>r?r@rArBrCrDrErFrGrHrIrJrKrLrMrNdeprecation_messagerss r__init__z+DPMSolverMultistepInverseScheduler.__init__s 6 ;; & &/A/CZ[ [  ++T[[-O-OQUQ\Q\QnQno pst tA  ; ;$3N3CDr#s  CWNa b  $m5==IDJ h & H>QY^YfYfgDJ o - C3H[chcpcpquvvDJ 1 1,-@ADJ%7OPTP^P^O_&`a aDJJ& #mmDKKQ?zz$"5"56 zz!d&9&9"9:  $,,/%))DLL2II D///43F3FF3N !$ !a a'''}'E)^,<rtjdd|jjz |dz} d | z }tj|jj@|zd|jj@dz |zzz ddj}||jjzj}tj2||ddgjtj4}ntjB|tj dtE||}d|j(|jz |j(|jz d z} tj2|| ggjtj4}tjF||_$tjJ|d \} } |tjL| }tjF|jO|tj|_(tE||_)dg|jjTz|_+d|_,d|_-|jHjOd|_$ycc}wcc}wcc}w)a Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. rr r7NleadingtrailingzY is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'.rU) in_sigmasror3T) return_indexr~rr_).r$ searchsortedfliprlr`rKitemr8noisiest_timesteprMrpr7roundrqastypeint64arangerNr arrayrgrkrF_convert_to_karras _sigma_to_t concatenater&rG_convert_to_exponentialrH_convert_to_betarIrJinterplenrrrmuniquesortrwrsror=rtrurv) rxror~ clipped_idxrs step_ratiorm log_sigmassigmare sigma_max_unique_indicess r set_timestepsz0DPMSolverMultistepInverseScheduler.set_timestepss((DMMA3)GIgIghmmo !%!@!@1!D{!R ;; ' ': 5 At557JQ7NOUUWX[Y[\aacjjkmkskst [[ ) )Y 60014:MPQ:QRJ1&9A&=>KRRTUXVXY^^`gghjhphpqI 11 1I [[ ) )Z 788;NNJ $"8"81"bggn56 )))477vU[[7Y#&y>   KK $ $%!" kknnU+ U"[ "["[s\ %\%:\*samplereturncb|j}|j^}}}|tjtjfvr|j }|j ||tj|z}|j}tj||jjd}tj|d|jj}|jd}tj|| ||z }|j ||g|}|j!|}|S)a{ "Dynamic thresholding: At each sampling step we set s to a certain percentile absolute pixel value in xt0 (the prediction of x_0 at timestep t), and if s > 1, then we threshold xt0 to the range [-s, s] and then divide by s. Dynamic thresholding pushes saturated pixels (those near -1 and 1) inwards, thereby actively preventing pixels from saturation at each step. We find that dynamic thresholding results in significantly better photorealism as well as better image-text alignment, especially when using very large guidance weights." https://huggingface.co/papers/2205.11487 r rW)r#max)rshaper$r&float64floatreshaperpprodabsquantiler`r@clamprA unsqueezerw)rxrr batch_sizechannelsremaining_dims abs_sampless r_threshold_samplez4DPMSolverMultistepInverseScheduler._threshold_sampleMs 06 - H~  6 6\\^F Hrww~7N,NOZZ\ NN:t{{'M'MST U KK 1$++66  KKNVaR+a/ HF~F5! rctjtj|d}||ddtjfz }tj|dk\dj dj |jddz }|dz}||}||}||z ||z z } tj | dd} d| z |z| |zz} | j|j} | S)Ng|=r)axisr)rr ) rprkmaximumnewaxiscumsumargmaxcliprr) rxrr log_sigmadistslow_idxhigh_idxlowhighwrs rrz.DPMSolverMultistepInverseScheduler._sigma_to_tosFF2::eU34 Jq"**}55))UaZq188a8@EE*JZJZ[\J]`aJaEbQ;!(#9_t , GGAq! Ug H , IIekk "rcr|jjr d|z }|}||fSd|dzdzdzz }||z}||fS)Nr rrU)r`rI)rxrrirjs r_sigma_to_alpha_sigma_tz:DPMSolverMultistepInverseScheduler._sigma_to_alpha_sigma_tsW ;; & &%iGG E1HqLS01GgoGrrct|jdr|jj}nd}t|jdr|jj}nd}||n|dj }||n|dj }d}t j dd|}|d|z z}|d|z z}||||z zz|z} | S)z6Constructs the noise schedule of Karras et al. (2022). sigma_minNrrrg@r )hasattrr`rrrrpr7) rxrrorrrhoramp min_inv_rho max_inv_rhorms rrz5DPMSolverMultistepInverseScheduler._convert_to_karrass 4;; , --II 4;; , --II!*!6IIbM The algorithm and model type are decoupled. You can use either DPMSolver or DPMSolver++ for both noise prediction and data prediction models. Args: model_output (`torch.Tensor`): The direct output from the learned diffusion model. sample (`torch.Tensor`): A current instance of a sample created by the diffusion process. Returns: `torch.Tensor`: The converted model output. rrNr /missing `sample` as a required keyword argumentrsrSPassing `timesteps` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`)r4rYr2)learned learned_ranger v_predictionflow_predictionzprediction_type given as zn must be one of `epsilon`, `sample`, `v_prediction`, or `flow_prediction` for the DPMSolverMultistepScheduler.rPz[ must be one of `epsilon`, `sample`, or `v_prediction` for the DPMSolverMultistepScheduler.) rpopr r r`rBr>rLrmr}rr?r) rxrrargskwargsrrrirjx0_predr2s rconvert_model_outputz7DPMSolverMultistepInverseScheduler.convert_model_outputsb:"$i!m47J1M >4y1}a !RSS   Z  ;; % %)K K{{**i7;;,,0LL#/2A2#6L DOO4#'#?#?#F !Gl$::gE,,8&,,> DOO4#'#?#?#F !F*W|-CC,,0AA++doo6 7\#99 / 0K0K/LM`` {{''009N[[ ' '+I I{{**i7;;,,0LL*1bqb51G*G,,8 DOO4#'#?#?#F !Gl$::gE,,> DOO4#'#?#?#F !L07V3CC / 0K0K/LMKK {{'' DOO4#'#?#?#F !Gg$55@009!Gg$55@N9Jrrnoiserc t|dkDr|dn|jdd}t|dkDr|dn|jdd}|t|dkDr|d}n td| tdd d | tdd d |j|j dz|j|j } }|j |\} }|j | \} } tj| tj|z } tj| tj| z } | | z }|jjd k(r*|| z |z| tj| d z z|zz }|S|jjdk(r)| | z |z|tj|d z z|zz }|S|jjdk(r||J|| z tj| z|z| dtjd|zz z|zz|tjd tjd|zz z|zz}|S|jjdk(rc|J| | z |zd|tj|d z zz|zz |tjtjd|zd z z|zz}S)a One step for the first-order DPMSolver (equivalent to DDIM). Args: model_output (`torch.Tensor`): The direct output from the learned diffusion model. sample (`torch.Tensor`): A current instance of a sample created by the diffusion process. Returns: `torch.Tensor`: The sample tensor at the previous timestep. rrNr prev_timesteprrrsrSrPassing `prev_timestep` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`r4r3rQrYrR@ rrr r rmr}rr$rkr`rBrrh)rxrrrrrrrrjsigma_srialpha_srllambda_shx_ts rdpm_solver_first_order_updatez@DPMSolverMultistepInverseScheduler.dpm_solver_first_order_updateIs*"$i!m47J1M#&t9q=QfjjRV6W >4y1}a !RSS   Z   $ ^   ;;t':;T[[=Y77@77@99W% '(::99W% '(:: x  ;; % % 6W$.'UYYr]S=P2QUa1aaC" ![[ ' '; 6W$.'UYYq\C=O2PT`1``C [[ ' '+< <$ $$7"UYYr]2f<a%))D1H"556,FGEJJsUYYrAv->'>??%GH  [[ ' '? :$ $$7"f,EIIaL3$678<GHEJJuyyQ'7#'=>>FG   rmodel_output_listc  t|dkDr|dn|jdd}t|dkDr|dn|jdd}|t|dkDr|d}n td| tddd | tddd |j|j dz|j|j |j|j dz } } }|j |\} }|j | \} } |j | \} } tj| tj|z }tj| tj| z }tj| tj| z }|d |d }}||z ||z }}||z }|d |z ||z z}}|jjdk(r|jjdk(rM|| z |z| tj| d z z|zz d| tj| d z zz|zz }|S|jjdk(rN|| z |z| tj| d z z|zz | tj| d z |z d zz|zz}S|jjdk(r|jjdk(rK| | z |z|tj|d z z|zz d|tj|d z zz|zz }|S|jjdk(rL| | z |z|tj|d z z|zz |tj|d z |z d z z|zz }S|jjdk(rv|J|jjdk(r|| z tj| z|z| dtjd|zz z|zzd| dtjd|zz zz|zz|tjd tjd |zz z|zz}|S|jjdk(r|| z tj| z|z| dtjd|zz z|zz| d tjd|zz d|zz d zz|zz|tjd tjd |zz z|zz}S|jjdk(r=|J|jjdk(r| | z |zd|tj|d z zz|zz |tj|d z z|zz |tjtjd|zd z z|zz}|S|jjdk(r| | z |zd|tj|d z zz|zz d|tj|d z |z d z zz|zz |tjtjd|zd z z|zz}S)a One step for the second-order multistep DPMSolver. Args: model_output_list (`List[torch.Tensor]`): The direct outputs from learned diffusion model at current and latter timesteps. sample (`torch.Tensor`): A current instance of a sample created by the diffusion process. Returns: `torch.Tensor`: The sample tensor at the previous timestep. r timestep_listNr rrrrSPassing `timestep_list` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`rrrr3r4r5rUr[rQrYrrRr)rrr r rmr}rr$rkr`rBrCrrh)rxrrrrrrrrjsigma_s0sigma_s1rialpha_s0alpha_s1rl lambda_s0 lambda_s1m0m1rh_0r0D0D1rs r(multistep_dpm_solver_second_order_updatezKDPMSolverMultistepInverseScheduler.multistep_dpm_solver_second_order_updates*$'t9q=QfjjRV6W #&t9q=QfjjRV6W >4y1}a !RSS  $ ^   $ ^  KK!+ , KK ( KK!+ ,$  77@!99(C(!99(C(99W% '(::IIh'%))H*== IIh'%))H*== "2&(9"(=BI%y9'<3 1WcBh27+B ;; % % 6{{&&*4x'61%))QB-#"56"<=W 1" (;<=BCr i((F2x'61%))QB-#"56"<=599aR=3#6!";c"ABbHIf ][[ ' '; 6{{&&*4x'61%))A,"45;<W ! s(:;4y1}a !RSS  $ ^   $ ^  KK!+ , KK ( KK!+ , KK!+ , 1 -8X 77@!99(C(!99(C(!99(C(99W% '(::IIh'%))H*== IIh'%))H*== IIh'%))H*== &r*,=b,ACTUWCXB*I ,A9yCX3q#'B Bh27+cBh27-Cd R27^t 4 4R"Wo$+ . ;; % % 68#v-eiimc12b89uyy!}s2a7#=>"DEuyy!}s2Q6!Q$>DEKL . #[[ ' '; 68#v-eiilS01R78uyy|c1Q6<=CDuyy|c1A5A=CDJK  [[ ' '+< <$ $$8#eiim3v=cEIIdQh$778B>?sUYYtax%88TAXFLMQSSTsUYYtax%8837BsQwSTnTWZZ[_aabEJJsUYYrAv->'>??%G H  rct|tjr%|j|jj }|j|k(j }t|dk(r t|jdz }||_ yt|dkDr|dj}||_ y|dj}||_ y)Nrr ) isinstancer$Tensorrwrsr~nonzerorrrv)rxrindex_candidatesr}s r_init_step_indexz3DPMSolverMultistepInverseScheduler._init_step_indexjs h -{{4>>#8#89H NNh6??A  A %T^^,q0J& ! "Q &)!,113J&*!,113J%rrvariance_noise return_dictc|j td|j|j||jt |j dz k(xrH|j jxs0|j jxrt |j dk}|jt |j dz k(xr0|j jxrt |j dk}|j||}t|j jdz D]!} |j| dz|j| <#||jd<|j jdvr0|.t|j||j |j" } n|j jdvr|} nd} |j jdk(s|j$dks|r|j'||| } nf|j jdk(s|j$dks|r|j)|j|| } n|j+|j|} |j$|j jkr|xj$dz c_|xj,dz c_|s| fSt/| S) a Predict the sample from the previous timestep by reversing the SDE. This function propagates the sample with the multistep DPMSolver. Args: model_output (`torch.Tensor`): The direct output from learned diffusion model. timestep (`int`): The current discrete timestep in the diffusion chain. sample (`torch.Tensor`): A current instance of a sample created by the diffusion process. generator (`torch.Generator`, *optional*): A random number generator. variance_noise (`torch.Tensor`): Alternative to generating noise with `generator` by directly providing the noise for the variance itself. Useful for methods such as [`CycleDiffusion`]. return_dict (`bool`): Whether or not to return a [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`. Returns: [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`: If return_dict is `True`, [`~schedulers.scheduling_utils.SchedulerOutput`] is returned, otherwise a tuple is returned where the first element is the sample tensor. NzaNumber of inference steps is 'None', you need to run 'set_timesteps' after creating the schedulerr rrr)rRrY) generatorr~rr) prev_sample)ror r}rrrsr`rErDrr!r=rtrBr rr~rrurrr rvr) rxrrrrrrrDlower_order_secondr+rrs rstepz'DPMSolverMultistepInverseScheduler.step}sD  # # +s  ?? "  ! !( +"__DNN0Ca0GG KK & & f4;;+H+H+eSQUQ_Q_M`ceMe __DNN 3a 7 7 wT[[=Z=Z w_bcgcqcq_ruw_w 00f0M t{{//!34 >A$($6$6q1u$=D  q ! >!-2 ;; % %)M MR`Rh ""i @S@S[g[m[mE[[ ' '+O O"EE ;; # #q (D,A,AA,EIZ<<\RX`e<fK [[ % % *d.C.Ca.GK]GGHZHZciqvGwKFFtGYGYbhFiK  4;;#;#; ;  ! !Q & ! A> !;77rc|S)a? Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. Args: sample (`torch.Tensor`): The input sample. Returns: `torch.Tensor`: A scaled input sample. )rxrrrs rscale_model_inputz4DPMSolverMultistepInverseScheduler.scale_model_inputs  roriginal_samplesrsc8|jj|j|j}|jjdk(rvt j |ra|jj|jt j}|j|jt j}n@|jj|j}|j|j}g}|D]x}||k(j}t|dk(rt|dz } n5t|dkDr|dj} n|dj} |j| z||j} t| jt|jkr=| jd} t| jt|jkr=|j!| \} } | |z| |zz} | S)Nrmpsrrr r)rmrwr~rtyper$is_floating_pointrsr&rrrr"flattenrrr)rxrrrsrmschedule_timesteps step_indicesrrr}rrirj noisy_sampless r add_noisez,DPMSolverMultistepInverseScheduler.add_noises'7'>'>FVF\F\]  " " ' '5 0U5L5LY5W!%!2!23C3J3JRWR_R_!2!` ! %5%<%GGI #$) !34q8 %&*-a0557 -a0557    + ,|$,,.%++%5%;%;!<<OOB'E%++%5%;%;!<< 77>"22Wu_D rc.|jjS)N)r`r8r|s r__len__z*DPMSolverMultistepInverseScheduler.__len__s{{...r)NN)333333?r*)NNT)+__name__ __module__ __qualname____doc__rname _compatiblesorderr rintstrrrrpndarrayrboolrzpropertyr}r$r~rrrrrrrrrrrr rrrrr IntTensorr'r)).0es00rr0r0NsDL%>>qAFF>L E$("%BF(",1"%+%"&$,116*/*/&)%*5\M'+ *1S/ S/S/ S/  S/  bjj$u+&= >? S/S/S/S/%*S/ S/S/S/ S/S/ $D>!S/"!)#S/$"$%S/&"$'S/(UO)S/*"+S/, }-S/./S/01S/S/j  U,U,U3PUP\P\K\E]U,p D0 ELLRWR^R^4TW\a\h\h.dgrAsN" // A1,XX !)4Xu/u/r