3jthSSKrSSKJr SSKJr SSKrSSKrSSKJ r J r SSK J r J r Jr SSKJr SS KJr \ "5(aSSKr\R*"\5r\"S S \ 55r"S S \\ 5rg)N) dataclass)Literal) ConfigMixinregister_to_config) BaseOutputis_scipy_availablelogging) randn_tensor)SchedulerMixinc8\rSrSr%Sr\R \S'Srg)FlowMatchLCMSchedulerOutput"a2 Output class for the scheduler's `step` function output. Args: prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images): Computed sample `(x_{t-1})` of previous timestep. `prev_sample` should be used as next model input in the denoising loop. prev_sampleN) __name__ __module__ __qualname____firstlineno____doc__torchTensor__annotations____static_attributes__rh/home/wildlama/miniconda3/lib/python3.13/site-packages/diffusers/schedulers/scheduling_flow_match_lcm.pyrr"srrc!>\rSrSrSr/rSr\S?S\S\ S\ S\ S \ S \S \S \ S \ S-S\ S-S\ S-S\ S-S\ SS\ \ S-S\ S4Sjj5r \S5r\S5r\S5rS@S\SS4SjjrS\ SS4SjrS\ \ S\SS4SjrSAS\R.S \ \R.-S!\R.S-S\R.4S"jjrS#\ \R.-S\ \R.-4S$jrS%\ S#\ S&\ \R6-\R8-S\ \R6-\R8-4S'jrS&\R6\R8-S\R6\R8-4S(jrSBS)\S-S*\\R>-S-S+\ \ S-S%\ S-S,\ \ S-SS4 S-jjr SAS \ \R8-S.\R8S-S\4S/jjr!S \ \R8-SS4S0jr"SCS1\R.S \ \R.-S\R.S2\RFS-S3\ S\$\%-4 S4jjr&S5\R8S)\S\R84S6jr'S5\R8S)\S\R84S7jr(SDS5\R8S)\S8\ S9\ S\R84 S:jjr)S%\ S#\ S&\ \R6-\R8-S\ \R6-\R8-4S;jr*S%\ S#\ S&\ \R6-\R8-S\ \R6-\R8-4S<jr+S\4S=jr,S>r-g)EFlowMatchLCMScheduler0ax LCM scheduler for Flow Matching. 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. shift (`float`, defaults to 1.0): The shift value for the timestep schedule. use_dynamic_shifting (`bool`, defaults to False): Whether to apply timestep shifting on-the-fly based on the image resolution. base_shift (`float`, defaults to 0.5): Value to stabilize image generation. Increasing `base_shift` reduces variation and image is more consistent with desired output. max_shift (`float`, defaults to 1.15): Value change allowed to latent vectors. Increasing `max_shift` encourages more variation and image may be more exaggerated or stylized. base_image_seq_len (`int`, defaults to 256): The base image sequence length. max_image_seq_len (`int`, defaults to 4096): The maximum image sequence length. invert_sigmas (`bool`, defaults to False): Whether to invert the sigmas. shift_terminal (`float`, defaults to None): The end value of the shifted timestep schedule. use_karras_sigmas (`bool`, defaults to False): Whether to use Karras sigmas for step sizes in the noise schedule during sampling. use_exponential_sigmas (`bool`, defaults to False): Whether to use exponential sigmas for step sizes in the noise schedule during sampling. use_beta_sigmas (`bool`, defaults to False): Whether to use beta sigmas for step sizes in the noise schedule during sampling. time_shift_type (`str`, defaults to "exponential"): The type of dynamic resolution-dependent timestep shifting to apply. scale_factors (`list[float]`, *optional*, defaults to `None`): It defines how to scale the latents at which predictions are made. upscale_mode (`str`, *optional*, defaults to "bicubic"): Upscaling method, applied if scale-wise generation is considered. r Nnum_train_timestepsshiftuse_dynamic_shifting base_shift max_shiftbase_image_seq_lenmax_image_seq_len invert_sigmasshift_terminaluse_karras_sigmasuse_exponential_sigmasuse_beta_sigmastime_shift_type) exponentiallinear scale_factors upscale_mode)nearestr/bilinearbicubic trilinearareaz nearest-exactcdURR(a[5(d [S5e[ URRURR URR /5S:a [S5eU S;a [S5e[R"SX[RS9SSS2R5n[R"U5R[RS9nUU- nU(dUU-SUS- U--- nUU-UlSUlSUlX lSUlXlXlURS5UlUR,SR/5UlUR,S R/5Ulg) 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.>r/r.z;`time_shift_type` must either be 'exponential' or 'linear'.dtypecpur)configr,r ImportErrorsumr+r* ValueErrornplinspacefloat32copyr from_numpyto timesteps _step_index _begin_index_shift _init_size_scale_factors _upscale_modesigmasitem sigma_min sigma_max)selfr!r"r#r$r%r&r'r(r)r*r+r,r-r0r1rFrMs r__init__FlowMatchLCMScheduler.__init__]s6 ;; & &/A/C/CZ[ [ KK//KK66KK11  A  "; ;Z[ [KK#6SUS]S]^_cac_cdiik $$Y/222G 00#V^qEAI+?'?@F"55  +)ii& R--/Q,,.rcUR$)z The value used for shifting. rIrQs rr"FlowMatchLCMScheduler.shifts {{rcUR$)zW The index counter for current timestep. It will increase 1 after each scheduler step. )rGrVs r step_index FlowMatchLCMScheduler.step_indexs rcUR$)za The index for the first timestep. It should be set from pipeline with `set_begin_index` method. rHrVs r begin_index!FlowMatchLCMScheduler.begin_indexs    rr]returncXlg)z Sets the begin index for the scheduler. This function should be run from pipeline before the inference. Args: begin_index (`int`, defaults to `0`): The begin index for the scheduler. Nr\)rQr]s rset_begin_index%FlowMatchLCMScheduler.set_begin_indexs (rcXlgNrU)rQr"s r set_shiftFlowMatchLCMScheduler.set_shifts rcXlX lg)z Sets scale factors for a scale-wise generation regime. Args: scale_factors (`list[float]`): The scale factors for each step. upscale_mode (`str`): Upscaling method. N)rKrL)rQr0r1s rset_scale_factors'FlowMatchLCMScheduler.set_scale_factorss,)rsampletimestepnoisec URRURURS9nURRS:Xav[ R "U5(a[URRUR[ RS9nURUR[ RS9nO@URRUR5nURUR5nURc!UVs/sHo`RXe5PM nnOHURbUR/URS-nOUR/URS-nXGR5n[UR5[UR5:a?URS5n[UR5[UR5:aM?X-SU- U--nU$s snf)a; Forward process in flow-matching Args: sample (`torch.FloatTensor`): The input sample. timestep (`torch.FloatTensor`): The current timestep in the diffusion chain. noise (`torch.FloatTensor`): The noise tensor. Returns: `torch.FloatTensor`: A scaled input sample. )devicer9mpsr8rr:?)rMrErnr9typeris_floating_pointrFrBr]index_for_timesteprYshapeflattenlen unsqueeze) rQrjrkrlrMschedule_timestepst step_indicessigmas r scale_noise!FlowMatchLCMScheduler.scale_noises,v}}FLLI ==   &5+B+B8+L+L!%!2!26== !2!V {{6== {FH!%!2!26==!A {{6==1H    #T\]T\q33AJT\L]L __ ( OO,x~~a/@@L!,,-q0AAL$,,.%++V\\!22OOB'E%++V\\!22#+!77 ^sHr{c2XRR-$rdr<r!)rQr{s r _sigma_to_t!FlowMatchLCMScheduler._sigma_to_ts{{6666rmurycURRS:XaURXU5$URRS:XaURXU5$g)Nr.r/)r<r-_time_shift_exponential_time_shift_linearrQrr{rys r time_shift FlowMatchLCMScheduler.time_shiftsS ;; & &- 7//1= = [[ ( (H 4**2a8 85rc\SU- nUSSURR- - nSX#- - nU$)a Stretches and shifts the timestep schedule to ensure it terminates at the configured `shift_terminal` config value. Reference: https://github.com/Lightricks/LTX-Video/blob/a01a171f8fe3d99dce2728d60a73fecf4d4238ae/ltx_video/schedulers/rf.py#L51 Args: t (`torch.Tensor` or `np.ndarray`): A tensor or numpy array of timesteps to be stretched and shifted. Returns: `torch.Tensor` or `np.ndarray`: A tensor or numpy array of adjusted timesteps such that the final value equals `self.config.shift_terminal`. r r:)r<r))rQry one_minus_z scale_factor stretched_ts rstretch_shift_to_terminal/FlowMatchLCMScheduler.stretch_shift_to_terminals="!e "2!dkk.H.H*HI ;56 rnum_inference_stepsrnrMrFcURR(aUc [S5eUb&Ub#[U5[U5:wa [S5eUb0Ub[U5U:wdUb[U5U:wa [S5eOUb [U5O [U5nXlUSLnU(a3[ R "U5R[ R5nUcfUcJ[ R"URUR5URUR5U5nXPRR- nO>[ R "U5R[ R5n[U5nURR(aURUSU5nO%URU-SURS- U--- nURR (aUR#U5nURR$(aUR'X1S9nOUURR((aUR+X1S9nO*URR,(aUR/X1S9n[0R2"U5R5[0RUS9nU(dX0RR-nO2[0R2"U5R5[0RUS9nURR6(aSSU- nX0RR-n[0R8"U[0R:"SUR<S 9/5nO5[0R8"U[0R>"SUR<S 9/5nXPl X0l!SUl"SUl#g) ar Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`, *optional*): 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. sigmas (`list[float]`, *optional*): Custom values for sigmas to be used for each diffusion step. If `None`, the sigmas are computed automatically. mu (`float`, *optional*): Determines the amount of shifting applied to sigmas when performing resolution-dependent timestep shifting. timesteps (`list[float]`, *optional*): Custom values for timesteps to be used for each diffusion step. If `None`, the timesteps are computed automatically. NzC`mu` must be passed when `use_dynamic_shifting` is set to be `True`z4`sigmas` and `timesteps` should have the same lengthzq`sigmas` and `timesteps` should have the same length as num_inference_steps, if `num_inference_steps` is providedrpr ) in_sigmasr)r9rn)rn)$r<r#r?rvrr@arrayastyperBrArrPrOr!rr"r)rr*_convert_to_karrasr+_convert_to_exponentialr,_convert_to_betarrDrEr(catonesrnzerosrFrMrGrH)rQrrnrMrrFis_timesteps_provideds r set_timesteps#FlowMatchLCMScheduler.set_timesteps$s4 ;; + + bc c  )"76{c)n, !WXX  *"s6{6I'I%#i.I > KK$$T^^4$$T^^4' !@!@@FXXf%,,RZZ8F"%f+  ;; + +__Rf5FZZ&(Aa60I,IJF ;; % %33F;F ;; ( (,,v,gF [[ / /11F1lF [[ ( (**V*eF!!&),,5==,P$!@!@@I((366U]]SY6ZI ;; $ $6\F!@!@@IYY 1V]](KLMFYY Afmm(LMNF"  rrxcUc URnX!:HR5n[U5S:aSOSn[X4R 55$)Nr r)rFnonzerorvintrN)rQrkrxindicesposs rrs(FlowMatchLCMScheduler.index_for_timestepsO  %!% %1::< w[U[R5(d[U[R5(a [ S5eUR (aKUR(a:[UR5[UR 5S-:wa [ S5eURb URcUR5SSUl URcURU5 UR[R5nUR URnUR URS-nX6U-- nUR (aUR(aUR"[UR 5:arURV s/sH(n [%UR UR"U -5PM* sn n [R&R(R+XURS9n[-UR.UUR0UR2S9n SU- U-Xz--n U=R"S- slU RUR25n U(dU 4$[5U S9$s sn f) a Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned model outputs (most often the predicted noise). Args: model_output (`torch.FloatTensor`): The direct output from learned diffusion model. timestep (`float`): The current discrete timestep in the diffusion chain. sample (`torch.FloatTensor`): A current instance of a sample created by the diffusion process. generator (`torch.Generator`, *optional*): A random number generator. return_dict (`bool`): Whether or not to return a [`~schedulers.scheduling_flow_match_lcm.FlowMatchLCMSchedulerOutput`] or tuple. Returns: [`~schedulers.scheduling_flow_match_lcm.FlowMatchLCMSchedulerOutput`] or `tuple`: If return_dict is `True`, [`~schedulers.scheduling_flow_match_lcm.FlowMatchLCMSchedulerOutput`] is returned, otherwise a tuple is returned where the first element is the sample tensor. zPassing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to `FlowMatchLCMScheduler.step()` is not supported. Make sure to pass one of the `scheduler.timesteps` as a timestep.r z]`_scale_factors` should have the same length as `timesteps` - 1, if `_scale_factors` are set.Nr)sizemode)rrnr9)r)rrr IntTensor LongTensorr?rKrLrvrFrJrYrrrErBrMrGroundnn functional interpolater rtrnr9r) rQrrkrjrrr{ sigma_nextx0_predrrlrs rstepFlowMatchLCMScheduler.steps"@ x % %(EOO44(E$4$455G    4#5#5#dnn:MQTUYUhUhQilmQm:mo  ?? "doo&=*//1!"5DO ?? "  ! !( +5==) DOO,[[1!45 <//   4#5#5#d&9&9"::X\XgXghXgPTd11$2B2BCdJKXgh((--99'SWSeSe9f MM>>--   :~0:3EE  A!nn\%7%78 > !*{CC'is9/Jrc[URS5(aURRnOSn[URS5(aURRnOSnUbUOUSR 5nUbUOUSR 5nSn[ R "SSU5nUSU- -nUSU- -nXXx- --U-n U $)a Construct the noise schedule as proposed in [Elucidating the Design Space of Diffusion-Based Generative Models](https://huggingface.co/papers/2206.00364). Args: in_sigmas (`torch.Tensor`): The input sigma values to be converted. num_inference_steps (`int`): The number of inference steps to generate the noise schedule for. Returns: `torch.Tensor`: The converted sigma values following the Karras noise schedule. rONrPr:rg@r )hasattrr<rOrPrNr@rA) rQrrrOrPrhoramp min_inv_rho max_inv_rhorMs rr(FlowMatchLCMScheduler._convert_to_karrass$ 4;; , , --II 4;; , , --II!*!6IIbMrsn ! A;;,,   H %  *   P /NKP /r