+hXSSKrSSKJrJrJrJr SSKrSSKrSSK J r J r SSK J r Jr SSKJrJrJr \"5(aSSKrS Sjr"S S \\ 5rg) N)ListOptionalTupleUnion) ConfigMixinregister_to_config) deprecateis_scipy_available)KarrasDiffusionSchedulersSchedulerMixinSchedulerOutputc &US:XaSnOUS:XaSnO[SU35e/n[U5H<nXP- nUS-U- nUR[SU"U5U"U5- - U55 M> [R "U[R S9$)a Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [0,1]. Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up to that part of the diffusion process. Args: num_diffusion_timesteps (`int`): the number of betas to produce. max_beta (`float`): the maximum beta to use; use values lower than 1 to prevent singularities. alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar. Choose from `cosine` or `exp` Returns: betas (`np.ndarray`): the betas used by the scheduler to step the model outputs cosinech[R"US-S- [R-S- 5S-$)NgMb?gT㥛 ?r)mathcospits h/home/james-whalen/.local/lib/python3.13/site-packages/diffusers/schedulers/scheduling_deis_multistep.py alpha_bar_fn)betas_for_alpha_bar..alpha_bar_fn;s-88QY%/$''9A=>!C Cexpc4[R"US-5$)Ng()rrrs rrr@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 * +  (!e. . S\"- R0@@@(KL, <<U]] 33rc.V\rSrSrSr\VVs/sHoR PM snnrSr\ SBS\ S\ S\ S\ S \ \RS \ S \ S \S \ S\ S\ S\ S\S\ \S\ \S\ \S\ \S\ \ S\ S\ S\S\ 4,Sjj5r\S5r\S5rSCS\ 4SjjrSDS \ S!\\ \R04S"\ \ 4S#jjrS$\R4S%\R44S&jrS'rS(rS)\R4S%\R44S*jrS)\R4S \ S%\R44S+jrSES)\R4S \ S,\ S-\ S%\R44 S.jjr SS/.S0\R4S$\R4S%\R44S1jjr!SS/.S0\R4S$\R4S%\R44S2jjr"SS/.S3\#\R4S$\R4S%\R44S4jjr$SS/.S3\#\R4S$\R4S%\R44S5jjr%SFS6jr&S7r'SGS0\R4S8\\ \R44S$\R4S9\S%\\(\)44 S:jjr*S$\R4S%\R44S;jr+S<\R4S=\R4S>\RXS%\R44S?jr-S@r.SAr/gs snnf)HDEISMultistepSchedulerNu `DEISMultistepScheduler` is a fast high order solver for diffusion ordinary differential equations (ODEs). 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 DEIS 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`): 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`. algorithm_type (`str`, defaults to `deis`): The algorithm type for the solver. lower_order_final (`bool`, defaults to `True`): Whether to use lower-order solvers in the final steps. Only valid for < 15 inference steps. 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. 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 Nnum_train_timesteps beta_startbeta_end beta_schedule trained_betas solver_orderprediction_type thresholdingdynamic_thresholding_ratiosample_max_valuealgorithm_type solver_typelower_order_finaluse_karras_sigmasuse_exponential_sigmasuse_beta_sigmasuse_flow_sigmas flow_shifttimestep_spacing steps_offsetuse_dynamic_shiftingtime_shift_typecURR(a[5(d [S5e[ URRURR URR /5S:a [S5eUb)[R"U[RS9Ul OUS:Xa*[R"X#U[RS9Ul OkUS:Xa4[R"US-US-U[RS9S-Ul O1US :Xa[U5Ul O[US UR35eS UR- Ul[R""UR S S 9Ul[R&"UR$5Ul[R&"SUR$- 5Ul[R,"UR(5[R,"UR*5- UlSUR$- UR$- S-UlS UlU S;a0U S;aUR5SS9 O[U S UR35eU S;a1U S;aUR5SS9 O[SU S UR35eSUl"S US- U[8RS9SSS2R;5n[R<"U5UlS/U-Ul S Ul!SUl"SUl#UR0RIS5Ulg)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.rlinear scaled_linear?rsquaredcos_cap_v2z is not implemented for ?rdim)deis) dpmsolverz dpmsolver++rP)r<)logrho)midpointheunbh1bh2rR)r=z solver type cpu)%configrAr ImportErrorsumr@r?r r$r%r&r*linspacer.NotImplementedError __class__alphascumprodalphas_cumprodsqrtalpha_tsigma_tloglambda_tsigmasinit_noise_sigmar num_inference_stepsnpcopy from_numpy timesteps model_outputslower_order_nums _step_index _begin_indexto)selfr2r3r4r5r6r7r8r9r:r;r<r=r>r?r@rArBrCrDrErFrGrms r__init__DEISMultistepScheduler.__init__s4 ;; & &/A/C/CZ[ [  ++T[[-O-OQUQ\Q\QnQno pst tA   $m5==IDJ h & >QY^YfYfgDJ o - C3H[chcpcpquvvDJ 1 1,-@ADJ%7OPTP^P^O_&`a aDJJ& #mmDKKQ?zz$"5"56 zz!d&9&9"9:  $,,/%))DLL2II D///43F3FF3N !$  )!==''v'>)^,<;;33 8S8SWd8d dd%'VVBZDKK " ;; ' ': 5 At{{>>BDWZ[D[\2"$!  [[ ) )Y 688=PST=TUJ1&9A&=>KRRTUYWYUYZ[^\^_ddfmmnpnvnvwI 11 1I [[ ) )Z 788;NNJ $++"A"A1zkRXXZ__ahhikiqiqrI NI;;//01JK A 3 33t7J7JJsRSVVF^ ;; ( (WWV_))+F,,v,gFSY!ZSY%"2"25"ESY!Z[aacI^^VBC[$9:AA"**MF [[ / /WWV_))+F11F1lFSY!ZSY%"2"25"ESY!Z[I^^VBC[$9:AA"**MF [[ ( (WWV_))+F**V*eFSY!ZSY%"2"25"ESY!Z[I^^VBC[$9:AA"**MF [[ ( ([[A (G(G$GI\_`I`aF6\FWWT[[33f<T[[E[E[^_E_ciDi@ijklomopuuwF++"A"AAGGII^^VBC[$9:AA"**MFYYy"))As6{*CVLFt221559L9LQ9OOTWWJ^^V\$:;BB2::NF&&v. )))477vU[[7Y#&y>   KK $ $%!"  kknnU+ I"[ "[ "[s \?]] samplereturncXURnURtp4nU[R[R4;aUR 5nUR X4[R"U5-5nUR5n[R"X`RRSS9n[R"USURRS9nURS5n[R"X*U5U- nUR "X4/UQ76nUR!U5nU$)aC "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 rN)r#max)rshaper$r&float64floatreshaperjprodabsquantilerYr:clampr; unsqueezerr)rsrr batch_sizechannelsremaining_dims abs_sampless r_threshold_sample(DEISMultistepScheduler._threshold_sampleAs 06 - ~  6 6\\^F rww~7N,NOZZ\ NN:{{'M'MST U KK 1$++66  KKNVR+a/ F~F5! rc[R"[R"US55nX2SS2[R4- n[R"US:SS9R SS9R URSS- S9nUS-nX%nX&nXs- Xx- - n [R "U SS5n SU - U-X--n U RUR5n U $)Ng|=r)axisr)rr ) rjremaximumnewaxiscumsumargmaxcliprr) rsrr log_sigmadistslow_idxhigh_idxlowhighwrs rr"DEISMultistepScheduler._sigma_to_tcsFF2::eU34 q"**}55))UaZq188a8@EE*JZJZ[\J]`aJaEbQ;!#_ , GGAq! Ug  , IIekk "rcvURR(a SU- nUnX#4$SUS-S-S-- nX-nX#4$)Nr rrK)rYrB)rsrrcrds r_sigma_to_alpha_sigma_t.DEISMultistepScheduler._sigma_to_alpha_sigma_t{sQ ;; & &%iGG E1HqLS01GoGrrc[URS5(aURRnOSn[URS5(aURRnOSnUbUOUSR 5nUbUOUSR 5nSn[ R "SSU5nUSU- -nUSU- -nXXx- --U-n U $)z6Constructs the noise schedule of Karras et al. (2022). sigma_minN sigma_maxrWrg@r )hasattrrYrritemrjr\) rsrrirrrhoramp min_inv_rho max_inv_rhorgs rr)DEISMultistepScheduler._convert_to_karrass 4;; , , --II 4;; , , --II!*!6IIbM4y1}a !RSS   Z   DOO,77> ;; & &) 3, 66'AG [[ ( (H 4"G [[ ( (N :&)??G [[ ( (,= =kk$//2G55G+DKK,G,G+HIWW  ;; # #,,W5G ;; % % /w..'9 9%&OP Prc,[U5S:aUSOURSS5n[U5S:aUSOURSS5nUc [U5S:aUSnO [S5eUb [SS S 5 Ub [SS S 5 URUR S-URUR pUR U5upUR U5up[R"U 5[R"U5- n [R"U 5[R"U5- n X- n URRS :Xa)X- U-U[R"U 5S - -U-- nU$[S5e)a One step for the first-order DEIS (equivalent to DDIM). Args: model_output (`torch.Tensor`): The direct output from the learned diffusion model. timestep (`int`): The current discrete timestep in the diffusion chain. prev_timestep (`int`): The previous discrete timestep in the diffusion chain. 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_timesteprrrmrrPassing `prev_timestep` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`rPrMr) rrr r rgrxrr$rerYr<rr])rsrrrrrrrdsigma_srcalpha_srflambda_shx_ts rdeis_first_order_update.DEISMultistepScheduler.deis_first_order_updates0"$i!m47J1M#&t9q=QfjjRV6W >4y1}a !RSS   Z   $ ^   ;;t':;T[[=Y77@77@99W% '(::99W% '(::   ;; % % /$.'UYYq\C=O2PT`1``C &&OP Prmodel_output_listc*[U5S:aUSOURSS5n[U5S:aUSOURSS5nUc [U5S:aUSnO [S5eUb [SSS 5 Ub [SSS 5 URUR S-URUR URUR S- pnUR U5upUR U5upUR U 5upUS US pXz- X- X- nnnURRS :XaBSnU"UUU5U"UUU5- nU"UUU5U"UUU5- nXU - UU --UU---nU$[S5e)au One step for the second-order multistep DEIS. 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 rrrrPassing `timestep_list` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`rrWrPcU[R"U5*[R"U5-S- -[R"U5[R"U5- - $)Nr rjre)rbcs rind_fnIDEISMultistepScheduler.multistep_deis_second_order_update..ind_fnsCRVVAYJ2Q67266!9rvvay;PQQrr rrr r rgrxrrYr<r])rsrrrrrrrdsigma_s0sigma_s1rcalpha_s0alpha_s1m0m1rho_trho_s0rho_s1rcoef1coef2rs r"multistep_deis_second_order_update9DEISMultistepScheduler.multistep_deis_second_order_updateKs($'t9q=QfjjRV6W #&t9q=QfjjRV6W >4y1}a !RSS  $ ^   $ ^  KK!+ , KK ( KK!+ ,$  77@!99(C!99(C"2&(9"(=B ' 183FH[vv ;; % % / R5&&1F6664RRE5&&1F6664RREh.;ebjHICJ%&OP Prc[U5S:aUSOURSS5n[U5S:aUSOURSS5nUc [U5S:aUSnO [S5eUb [SSS 5 Ub [SSS 5 URUR S-URUR URUR S- URUR S- 4upxpUR U5upUR U5upUR U 5upUR U 5upUS US US nnnX{- X- X- X- 4unnnnURRS:XacSnU"UUUU5U"UUUU5- nU"UUUU5U"UUUU5- nU"UUUU5U"UUUU5- nXU - UU--UU--UU---nU$[S5e)ap One step for the third-order multistep DEIS. 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 diffusion process. Returns: `torch.Tensor`: The sample tensor at the previous timestep. rrNr rrrrrrrWrrPcLU[R"U5[R"U5[R"U5- S--[R"U5[R"U5-- [R"U5-[R"U5S--S[R"U5-- S--n[R"U5[R"U5- [R"U5[R"U5- -nXE- $)Nr rr)rrrd numerator denominators rrHDEISMultistepScheduler.multistep_deis_third_order_update..ind_fnsFF1IRVVAY!6!:;ffQi"&&)+,ffQi ffQi1n%"&&)m $    "vvay266!94RVVAY9NO  ..rrr)rsrrrrrrrdrrsigma_s2rcrralpha_s2rrm2rrrrho_s2rrrcoef3rs r!multistep_deis_third_order_update8DEISMultistepScheduler.multistep_deis_third_order_updatesP*$'t9q=QfjjRV6W #&t9q=QfjjRV6W >4y1}a !RSS  $ ^   $ ^  KK!+ , KK ( KK!+ , KK!+ , 1 -8 77@!99(C!99(C!99(C&r*,=b,ACTUWCXB         ) %vvv ;; % % / /5&&&9F66SY[arr!r7rnrorrrrpr) rsrrrr"r>lower_order_secondr+r%s rstepDEISMultistepScheduler.step s 6  # # +s  ?? "  ! !( +__DNN 3a 7 7 wT[[=Z=Z w_bcgcqcq_ruw_w __DNN 3a 7 7 wT[[=Z=Z w_bcgcqcq_ruw_w 000M t{{//!34A$($6$6q1u$=D  q !5!-2 ;; # #q (D,A,AA,EIZ66|6SK [[ % % *d.C.Ca.GK]AA$BTBT]cAdK@@ASAS\b@cK  4;;#;#; ;  ! !Q & ! A> !;77rcU$)z 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. )rsrrrs rscale_model_input(DEISMultistepScheduler.scale_model_inputJs  roriginal_samplesnoisermc*URRURURS9nURRS:Xav[ R "U5(a[URRUR[ RS9nURUR[ RS9nO@URRUR5nURUR5nURc!UVs/sHo`RXe5PM nnOHURbUR/URS-nOUR/URS-nXGR5n[UR5[UR5:a?URS5n[UR5[UR5:aM?UR!U5upX-X--n U $s snf)NrmpsrrrW)rgrrrrtyper$is_floating_pointrmr&r|rrxrflattenrrr) rsr-r.rmrgrr step_indicesrrcrd noisy_sampless r add_noise DEISMultistepScheduler.add_noiseZs'7'>'>FVF\F\]  " " ' '5 0U5L5LY5W5W!%!2!23C3J3JRWR_R_!2!` ! %5%<%2W_D _sHc.URR$N)rYr2rws r__len__DEISMultistepScheduler.__len__|s{{...r)rqrprcr_rar*rhrfrornrirdrgrm)ig-C6?g{Gz?rINrrFgףp= ?rMrPrRTFFFFrMr\rFr)r)NN)333333?r<r9)T)0__name__ __module__ __qualname____firstlineno____doc__r name _compatiblesorderr intrstrrrjndarrayboolrtpropertyrxr|rrr$rrrrrrrrrrrrrrrr rrr'r+ IntTensorr6r:__static_attributes__).0es00rr0r0NsV1f%>>$=qFF$=>L E$("%.2(",1"%$#"&,116*/*/&) *%*,/L, L,L, L,  L,   + L,L,L,L,%*L, L,L,L, L,$D>L, !)!L,""$#L,$"$%L,&UO'L,()L,*+L,,#-L,./L,L,\  !!(3(hlQ,#&Q,05c5<<6G0HQ,U]^cUdQ,h D0 ELLRWR^R^4TW\a\h\h.dgrTsN$ // A1XX !)4Xo /^[o /r