wiPddlZddlZddlmZddlmZmZmZmZddl Z ddl Z ddl m Z ddlmZmZddlmZdd lmZmZeGd d eZ ddZGddeeZdS)N) dataclass)ListOptionalTupleUnion) integrate) ConfigMixinregister_to_config) BaseOutput)KarrasDiffusionSchedulersSchedulerMixincLeZdZUdZejed<dZeejed<dS)LMSDiscreteSchedulerOutputaq Output class for the scheduler's `step` function output. Args: prev_sample (`torch.Tensor` 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. pred_original_sample (`torch.Tensor` of shape `(batch_size, num_channels, height, width)` for images): The predicted denoised sample `(x_{0})` based on the model output from the current timestep. `pred_original_sample` can be used to preview progress or for guidance. prev_sampleNpred_original_sample) __name__ __module__ __qualname____doc__torchTensor__annotations__rr|/root/.openclaw/workspace/chatterbox_venv_py311/lib/python3.11/site-packages/diffusers/schedulers/scheduling_lms_discrete.pyrrsD  37(5<077777rr+?cosinec F|dkrd}n|dkrd}ntd|g}t|D]J}||z }|dz|z }|td||||z z |Kt j|tjS)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 rc\tj|dzdz tjzdz dzS)NgMb?gT㥛 ?r )mathcospits r alpha_bar_fnz)betas_for_alpha_bar..alpha_bar_fnIs,8QY%/$'9A=>>!C Crexpc0tj|dzS)Ng()r"r(r%s rr'z)betas_for_alpha_bar..alpha_bar_fnNs8AI&& &rz"Unsupported alpha_transform_type: r dtype) ValueErrorrangeappendminrtensorfloat32)num_diffusion_timestepsmax_betaalpha_transform_typer'betasit1t2s rbetas_for_alpha_barr90s.x'' D D D D  & & ' ' ' 'T>RTTUUU E * + +MM ( (!e. . S\\"-- R0@0@@@(KKLLLL <U] 3 3 33rcbeZdZdZdeDZdZe d4d ede de de de e e jee fde ede de defdZedZedZedZd5defdZdejde e ejfdejfdZd Zd6d!ed"e e ejffd#Zd6d$Zd%Zd&Zd'ejdejfd(Z d7d+ejde e ejfdejd,ed-ede e!e"ff d.Z#d/ejd0ejd1ejdejfd2Z$d3Z%dS)8LMSDiscreteScheduleru A linear multistep scheduler for discrete beta schedules. 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` or `scaled_linear`. trained_betas (`np.ndarray`, *optional*): Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`. 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}. 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). 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. cg|] }|j Sr)name).0es r zLMSDiscreteScheduler.}s>>>qAF>>>rr -C6?{Gz?linearNFepsilonlinspacernum_train_timesteps beta_startbeta_end beta_schedule trained_betasuse_karras_sigmasprediction_typetimestep_spacing steps_offsetc |&tj|tj|_n|dkr(tj|||tj|_nk|dkr1tj|dz|dz|tjdz|_n4|dkrt ||_nt |d|jd|jz |_tj |jd |_ tj d |j z |j z dz} tj | ddd d ggtj} tj| |_d|_||_||dg|_d|_d|_d|_|jd|_dS)Nr*rD scaled_linear?r squaredcos_cap_v2z is not implemented for ?r)dimr Fcpu)rr0r1r5rFr9NotImplementedError __class__alphascumprodalphas_cumprodnparray concatenateastype from_numpysigmasnum_inference_stepsrL set_timesteps derivativesis_scale_input_called _step_index _begin_indexto) selfrGrHrIrJrKrLrMrNrOrcs r__init__zLMSDiscreteScheduler.__init__s  $m5=IIIDJJ h & & H>QY^YfgggDJJ o - - C3H[chcpqqquvvDJJ 1 1 1,-@AADJJ%&`&`PTP^&`&`aa aDJ& #mDKQ???A 33t7JJsRSS"u 566==bjII&v.. $( !2 .555%*" knnU++ rc|jjdvr|jS|jdzdzdzS)N)rFtrailingr r rR)configrNrcmaxrks rinit_noise_sigmaz%LMSDiscreteScheduler.init_noise_sigmasH ; '+C C C;??$$ $ !!Q&*s22rc|jS)zg The index counter for current timestep. It will increase 1 after each scheduler step. )rhrqs r step_indexzLMSDiscreteScheduler.step_indexs rc|jS)zq The index for the first timestep. It should be set from pipeline with `set_begin_index` method. rirqs r begin_indexz LMSDiscreteScheduler.begin_indexs   rrwc||_dS)z Sets the begin index for the scheduler. This function should be run from pipeline before the inference. Args: begin_index (`int`): The begin index for the scheduler. Nrv)rkrws rset_begin_indexz$LMSDiscreteScheduler.set_begin_indexs(rsampletimestepreturnc|j|||j|j}||dzdzdzz }d|_|S)a Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. Args: sample (`torch.Tensor`): The input sample. timestep (`float` or `torch.Tensor`): The current timestep in the diffusion chain. Returns: `torch.Tensor`: A scaled input sample. Nr r rRT)rt_init_step_indexrcrg)rkrzr{sigmas rscale_model_inputz&LMSDiscreteScheduler.scale_model_inputsR ? "  ! !( + + + DO,E1HqLS01%)" rcfd}tj|jjdzdd}|S)z Compute the linear multistep coefficient. Args: order (): t (): current_order (): cd}tD]A}|kr ||j|z z jz j|z z z z}B|S)NrT)r-rc)tauprodk current_orderorderrkr&s rlms_derivativez@LMSDiscreteScheduler.get_lms_coefficient..lms_derivativestD5\\ k k A%%t{1q511dk!mBS6TW[WbcdghchWi6ijjKrr rB)epsrelr)rquadrc)rkrr&rrintegrated_coeffs```` rget_lms_coefficientz(LMSDiscreteScheduler.get_lms_coefficientsj        %>.$+a.$+VWZ[V[J\eijjjklmrrddevicec 8|_jjdkrLtjdjjdz |tjddd}n>jjdkrjjjz}tjd||z ddd tj}|jj z }njjdkr|jjjz }tjjjd|   tj}|dz}ntjjd tj djz jz d z}tj|tj|tjdt#||}jjr7| }tj fd |D}tj|d gg tj}t+j||_t+j||_d_d_jd_g_dS)a Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. rFrr r*NrVleadingrnzY is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'.rR) in_sigmasc<g|]}|Sr) _sigma_to_t)r>r log_sigmasrks rr@z6LMSDiscreteScheduler.set_timesteps..#s)!Z!Z!Z%$"2"25*"E"E!Z!Z!ZrrW)rrX)rdrorNr^rFrGr1copyarangeroundrarOr,r_r]loginterplenrL_convert_to_karrasr`rrbrjrc timestepsrhrirf)rkrdrr step_ratiorcrs` @rrez"LMSDiscreteScheduler.set_timestepss$7  ; ': 5 5 At{'F'JL_gigqrrr"dff I[ )Y 6 68D>F!Z!Z!Z!Z!ZSY!Z!Z!Z[[I#0077 CC&v..111@@ ))4477v7FF knnU++ rc||j}||k}t|dkrdnd}||S)Nr r)rnonzeroritem)rkr{schedule_timestepsindicesposs rindex_for_timestepz'LMSDiscreteScheduler.index_for_timestep0sW  %!% %1::<< w<#;;t~'<==#66x@@D   #0D   rctjtj|d}||ddtjfz }tj|dkdd|jddz }|dz}||}||}||z ||z z } tj| dd} d| z |z| |zz} | |j} | S)Ng|=r)axisr )rpr ) r^rmaximumnewaxiscumsumargmaxclipshapereshape) rkrr log_sigmadistslow_idxhigh_idxlowhighwr&s rrz LMSDiscreteScheduler._sigma_to_tHsF2:eU3344 Jqqq"*}55)UaZq11188a8@@EE*JZ[\J]`aJaEbbQ;!(#9_t , GAq!  Ug H , IIek " "rrc|d}|d}d}tjdd|j}|d|z z}|d|z z}||||z zz|z}|S)z6Constructs the noise schedule of Karras et al. (2022).rVrg@r )rr^rFrd) rkr sigma_min sigma_maxrhoramp min_inv_rho max_inv_rhorcs rrz'LMSDiscreteScheduler._convert_to_karras`s%R=--// $Q<,,.. {1a!9::AG, AG,  k(A BBsJ rT model_outputr return_dictc \jstjdj|jj}jjdkr |||zz }n_jjdkr|| |dzdzdzz z||dzdzz z}n0jjdkr|}ntd jjd ||z |z }j |tj krj d tjdzfd tD} |td t!| t#j Dz} xjdz c_|s| fSt'| |S)aa 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.Tensor`): The direct output from learned diffusion model. timestep (`float` or `torch.Tensor`): The current discrete timestep in the diffusion chain. sample (`torch.Tensor`): A current instance of a sample created by the diffusion process. order (`int`, defaults to 4): The order of the linear multistep method. return_dict (`bool`, *optional*, defaults to `True`): 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. zThe `scale_model_input` function should be called before `step` to ensure correct denoising. See `StableDiffusionPipeline` for a usage example.NrE v_predictionr r rRrzzprediction_type given as z, must be one of `epsilon`, or `v_prediction`rcHg|]}j|Sr)rrt)r> curr_orderrrks rr@z-LMSDiscreteScheduler.step..s.rrrWad..udozRRrrrrc3&K|] \}}||zV dSrr)r>coeff derivatives r z,LMSDiscreteScheduler.step..s<# # #45*EJ # # # # # # r)rr)rgwarningswarnrtr~rcrorMr,rfr.rpopr/r-sumzipreversedrhr) rkrr{rzrrrrr lms_coeffsrs ` ` rstepzLMSDiscreteScheduler.stepms<)  ME    ? "  ! !( + + + DO, ; &) 3 3#)EL,@#@ [ (N : :#/E6UAX\c !)kXlmmmmroriginal_samplesnoiserc`j|j|j}|jjdkrft j|rRj|jt j||jt j}n9j|j||j}j fd|D}n4j j g|j dz}nj g|j dz}|| }t|j t|j kr?|d}t|j t|j k?|||zz}|S)N)rr+mpsr*c<g|]}|Sr)r)r>r&rrks rr@z2LMSDiscreteScheduler.add_noise..s*^^^qD33A7IJJ^^^rrrV)rcrjrr+typeris_floating_pointrr1rwrtrflattenr unsqueeze) rkrrrrc step_indicesr noisy_samplesrs ` @r add_noisezLMSDiscreteScheduler.add_noises'7'>FVF\]]  " '5 0 0U5LY5W5W 0!%!2!23C3JRWR_!2!`!` ! %5%$=>>>L E$("%BF,1( *),), ),), ),  ),  bj$u+&= >? ),$D>),),),),),),),V33X3  X !!X!((3(((( eU\FY@Z_d_k0   ,222eCDU>V2222j # # # #1110 EL U\    $ KnKnlKnu|+,Kn Kn  Kn  Kn )50 1KnKnKnKn\,|<   B/////rr;)rr)r"r dataclassesrtypingrrrrnumpyr^rscipyrconfiguration_utilsr r utilsr scheduling_utilsrrrr9r;rrrrsK !!!!!!//////////// AAAAAAAAGGGGGGGG 8888888 8(!)4)4)4)4XA/A/A/A/A/>;A/A/A/A/A/r