wi^ddlZddlmZmZmZmZddlZddlZddl Z ddl m Z m Z ddl mZmZmZGddZGd d Z dd ZGddee ZdS)N)ListOptionalTupleUnion) ConfigMixinregister_to_config)KarrasDiffusionSchedulersSchedulerMixinSchedulerOutputc6eZdZdZddZedZdZdS)BatchedBrownianTreezGA wrapper around torchsde.BrownianTree that enables batches of entropy.Nc |\|_dtj||(tjddg}d|_ t||j dksJdn#t$r |g}d|_YnwxYwfd|D|_ dS)Nw0rlTFc<g|]}tjfd|iS)entropy)torchsde BrownianTree).0skwargst0t1rs }/root/.openclaw/workspace/chatterbox_venv_py311/lib/python3.11/site-packages/diffusers/schedulers/scheduling_dpmsolver_sde.py z0BatchedBrownianTree.__init__..)s6[[[QRh+BBLLLVLL[[[) sortsigngettorch zeros_likerandintitembatchedlenshape TypeErrortrees)selfxrrseedrrs `` `@r__init__zBatchedBrownianTree.__init__s IIb"--B ZZe.q11 2 2 <=Ir227799D  !t99 ****ABB ! ! !6D DLLL !\[[[[[[VZ[[[ s>(B''B>=B>c"||kr||dfn||dfS)Nr )abs rrzBatchedBrownianTree.sort+s EE1ayy1bz1rc|\}tjfd|jD|j|zz}|jr|n|dS)Nc(g|]}|Sr0r0)rtreerrs rrz0BatchedBrownianTree.__call__..1s#===$b"===rr)rr!stackr)rr%)r*rrrws `` r__call__zBatchedBrownianTree.__call__/sgyyR(( B K=====$*=== > >$)dBR SL*qqad*rN)__name__ __module__ __qualname____doc__r- staticmethodrr8r0rrrrs\QQ \ \ \ \22\2+++++rrc&eZdZdZddfdZdZdS)BrownianTreeNoiseSamplerasA noise sampler backed by a torchsde.BrownianTree. Args: x (Tensor): The tensor whose shape, device and dtype to use to generate random samples. sigma_min (float): The low end of the valid interval. sigma_max (float): The high end of the valid interval. seed (int or List[int]): The random seed. If a list of seeds is supplied instead of a single integer, then the noise sampler will use one BrownianTree per batch item, each with its own seed. transform (callable): A function that maps sigma to the sampler's internal timestep. Nc|Sr9r0)r+s rz!BrownianTreeNoiseSampler.Dsqrc||_|tj||tj|}}t|||||_dSr9) transformr! as_tensorrr5)r*r+ sigma_min sigma_maxr,rDrrs rr-z!BrownianTreeNoiseSampler.__init__DsX" : :;;T^^EO\eLfLf=g=gB'2r488 rc|tj||tj|}}|||||z z Sr9)rDr!rEr5abssqrt)r*sigma sigma_nextrrs rr8z!BrownianTreeNoiseSampler.__call__Isi 6 677XbHcHc9d9dByyR  BG==??#7#7#9#999r)r:r;r<r=r-r8r0rrr@r@5sJ  6:[[9999 :::::rr@+?cosinec F|dkrd}n|dkrd}ntd|g}t|D]J}||z }|dz|z }|td||||z z |Kt j|tjS)a Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [0,1]. Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up to that part of the diffusion process. Args: num_diffusion_timesteps (`int`): the number of betas to produce. max_beta (`float`): the maximum beta to use; use values lower than 1 to prevent singularities. alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar. Choose from `cosine` or `exp` Returns: betas (`np.ndarray`): the betas used by the scheduler to step the model outputs rNc\tj|dzdz tjzdz dzS)NgMb?gT㥛 ?r)mathcospits r alpha_bar_fnz)betas_for_alpha_bar..alpha_bar_fnhs,8QY%/$'9A=>>!C Crexpc0tj|dzS)Ng()rQrWrTs rrVz)betas_for_alpha_bar..alpha_bar_fnms8AI&& &rz"Unsupported alpha_transform_type: r dtype) ValueErrorrangeappendminr!tensorfloat32)num_diffusion_timestepsmax_betaalpha_transform_typerVbetasirt2s rbetas_for_alpha_barrgOs.x'' D D D D  & & ' ' ' 'T>RTTUUU E * + +MM ( (!e. . S\\"-- R0@0@@@(KKLLLL <U] 3 3 33rceZdZdZdeDZdZe d6d ede de de de e e jee fde de ede ede defdZd7dZdZedZedZedZd8defdZdejd e e ejfd!ejfd"Z d9d#ed$e e ejfd e efd%Zd&Zd'Zd(ejd!ejfd)Z ed*Z! d:d-e eje jfd e e ejfde eje jfd.ed/e d!e e"e#ff d0Z$d1ejd2ejd3ejd!ejfd4Z%d5Z&dS);DPMSolverSDEScheduleru DPMSolverSDEScheduler implements the stochastic sampler from the [Elucidating the Design Space of Diffusion-Based Generative Models](https://huggingface.co/papers/2206.00364) paper. 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 0.00085): The starting `beta` value of inference. beta_end (`float`, defaults to 0.012): 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`. 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). 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}. noise_sampler_seed (`int`, *optional*, defaults to `None`): The random seed to use for the noise sampler. If `None`, a random seed is generated. 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 Sr0)name)res rrz DPMSolverSDEScheduler.s>>>qAF>>>rr_QK?~jt?linearNepsilonFlinspacernum_train_timesteps beta_startbeta_end beta_schedule trained_betasprediction_typeuse_karras_sigmasnoise_sampler_seedtimestep_spacing steps_offsetc |&tj|tj|_n|dkr(tj|||tj|_nk|dkr1tj|dz|dz|tjdz|_n4|dkrt ||_nt |d|jd|jz |_tj |jd |_ | |d|||_ d|_ ||_d|_d|_|jd |_dS) NrYrp scaled_linear?rsquaredcos_cap_v2z is not implemented for ?r)dimcpu)r!r_r`rdrrrgNotImplementedError __class__alphascumprodalphas_cumprod set_timestepsry noise_samplerrz _step_index _begin_indexsigmasto) r*rsrtrurvrwrxryrzr{r|s rr-zDPMSolverSDEScheduler.__init__sL  $m5=IIIDJJ h & & H>QY^YfgggDJJ o - - C3H[chcpqqquvvDJJ 1 1 1,-@AADJJ%&`&`PTP^&`&`aa aDJ& #mDKQ??? .6IJJJ!2!"4 knnU++ rc||j}||k}t|dkrdnd}||S)Nr r) timestepsnonzeror&r$)r*timestepschedule_timestepsindicesposs rindex_for_timestepz(DPMSolverSDEScheduler.index_for_timestepsW  %!% %1::<< w<#;;t~'<==#66x@@D   #0D   rc|jjdvr|jS|jdzdzdzS)N)rrtrailingrr r)configr{rmaxr*s rinit_noise_sigmaz&DPMSolverSDEScheduler.init_noise_sigmasH ; '+C C C;??$$ $ !!Q&*s22rc|jS)zg The index counter for current timestep. It will increase 1 after each scheduler step. )rrs r step_indexz DPMSolverSDEScheduler.step_indexs rc|jS)zq The index for the first timestep. It should be set from pipeline with `set_begin_index` method. rrs rrz!DPMSolverSDEScheduler.begin_indexs   rrc||_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. Nr)r*rs rset_begin_indexz%DPMSolverSDEScheduler.set_begin_indexs(rsamplerreturnc|j|||j|j}|jr|n|j}||dzdzdzz }|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 (`int`, *optional*): The current timestep in the diffusion chain. Returns: `torch.Tensor`: A scaled input sample. Nrr r)rrrstate_in_first_ordermid_point_sigma)r*rrrK sigma_inputs rscale_model_inputz'DPMSolverSDEScheduler.scale_model_inputs_& ? "  ! !( + + + DO,#8Reed>R KNQ.367 rnum_inference_stepsrc |_|p jj}jjdkr=t jd|dz |t ddd}njjdkrw|jz}t jd||z ddd t }|jj z }njjdkrc|jz }t j|d|   t }|dz}ntjjd t j djz jz d z}t j|t j|t jdt#||}jjr7| }t j fd |D}|}t j|d gg tj}t/j||}t/j|dd|ddd|ddg_t/j|}t/j|}t/j|dd|dddg}||ddd<t;|dr'||t.j_n||_d_ d_!d_"d_#jd_d_$dS)a Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. rrrr rYNr/leadingrzY is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'.r) in_sigmasc<g|]}|Sr0 _sigma_to_trrK log_sigmasr*s rrz7DPMSolverSDEScheduler.set_timesteps..Js)!Z!Z!Z%$"2"25*"E"E!Z!Z!Zrg)rrmpsr)%rrrsr{nprrfloatcopyarangeroundastyper|r[arrayrloginterpr&ry_convert_to_karras_second_order_timesteps concatenater`r! from_numpyrcatrepeat_interleaverstr startswithrrrrrr) r*rrrsr step_ratiorsecond_order_timestepsrs ` @rrz#DPMSolverSDEScheduler.set_timestepss$7 1TT[5T ; ': 5 5 A':Q'>@S[`aaabfbfdfbfgllnnII [ )Y 6 6,0HHJ1&9::ZGNNPPQUQUSUQUV[[]]ddejkkI 1 1II [ )Z 7 7,t/GGJ#6J;GGNNPPUUWW^^_deeI NII;/KKK A 33t7JJsRSSVF^^ 9bi3v;;&?&?HH ; ( \,,v,>>F!Z!Z!Z!Z!ZSY!Z!Z!Z[[I!%!=!=fj!Q!Q#0077 CC!&)),,F,;;iVAbD\-K-KA-N-NPVWYWZWZP[ \]] $Y// !&!12H!I!IIy!}im.M.Ma.P.PQRR 0 !$Q$ v;; ! !% ( ( 9&\\& \FFDNN&\\\88DN # knnU++ !rcd}d}d}||}tj|}|dd||zz}||} tjfd| D} | S)Nc,tj| Sr9)rrW_ts rsigma_fnz?DPMSolverSDEScheduler._second_order_timesteps..sigma_fngs62#;; rc,tj| Sr9)rr_sigmas rt_fnz;DPMSolverSDEScheduler._second_order_timesteps..t_fnjsF6NN? "rrr/c<g|]}|Sr0rrs rrzADPMSolverSDEScheduler._second_order_timesteps..rs)\\\ed..ujAA\\\r)rdiffr) r*rrrrmidpoint_ratiorU delta_time t_proposed sig_proposedrs ` ` rrz-DPMSolverSDEScheduler._second_order_timestepsfs    # # # DLLWQZZ ssVj>99 x ++ H\\\\\|\\\]] rctjtj|d}||ddtjfz }tj|dkdd|jddz }|dz}||}||}||z ||z z } tj| dd} d| z |z| |zz} | |j} | S)Ng|=r)axisr)rr ) rrmaximumnewaxiscumsumargmaxclipr'reshape) r*rKr log_sigmadistslow_idxhigh_idxlowhighr7rUs rrz!DPMSolverSDEScheduler._sigma_to_tvsF2: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).r/rg@r )r$rrrr) r*rrFrGrhoramp min_inv_rho max_inv_rhors rrz(DPMSolverSDEScheduler._convert_to_karrass%R=--// $Q<,,.. {1a!9::AG, AG,  k(A BBsJ rc|jduSr9)rrs rrz*DPMSolverSDEScheduler.state_in_first_orders{d""rTr model_output return_dicts_noisec8|j|||j]|j|jdk|j}}t ||||j|_dtj dtj fd}dtj dtj fd} |j r(|j|j} |j|jdz} n'|j|jdz } |j|j} d } | | | | }} || z }| || zz}|j j d kr|j r| n ||}|||zz }n|j j d kr3|j r| n ||}|| |d zdzd zz z||d zdzz z}n<|j j d krtdtd|j j d| dkr||z | z }| | z }|||zz}n|j r|}n|j}|| }||}t ||d z|d z|d zz z|d zz d z}|d z|d zz d z}| |}|||| z |z| |z |zz }|||| |||z|zz}|j r||_|||_nd|_d|_|xjdz c_|s|fSt'|S)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.Tensor` or `np.ndarray`): The direct output from learned diffusion model. timestep (`float` or `torch.Tensor`): The current discrete timestep in the diffusion chain. sample (`torch.Tensor` or `np.ndarray`): A current instance of a sample created by the diffusion process. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~schedulers.scheduling_utils.SchedulerOutput`] or tuple. s_noise (`float`, *optional*, defaults to 1.0): Scaling factor for noise added to the sample. 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. NrrrcN|Sr9)negrWrs rrz,DPMSolverSDEScheduler.step..sigma_fns6688<<>> !rrcN|Sr9)rrrs rrz(DPMSolverSDEScheduler.step..t_fns::<<##%% %rr rrq v_predictionrrz+prediction_type not implemented yet: samplezprediction_type given as z, must be one of `epsilon`, or `v_prediction`) prev_sample)rrrrr^rr@rzr!rrrrxrr[rexpm1rrr )r*rrrrr min_sigma max_sigmarrrKrLrrUt_nextrrrpred_original_sample derivativedtr sigma_fromsigma_tosigma_up sigma_down ancestral_ts rstepzDPMSolverSDEScheduler.steps: ? "  ! !( + + +   %#';t{Q#?#C#C#E#Et{GXGXyI!9&)YX\Xo!p!pD  " "%, " " " " & &%, & & & &  $ 6K0ET_q%89JJK! 34ET_5JDKKj!1!16aZ n44  ; &) 3 3#'#<V%%((:BVBVK#)K,,F#F [ (N : :#'#<V%%((:BVBVK#/K<;PQ>TUCUZ]B]3]#^+q.1,-$ [ (H 4 4%&STT TuDK,Guuu  ?? #775@Je#B :?2KK( %#!!Jx''H8hkZ]Xq[5P&QT^`aTa&afi%ijjH"A+! 3;J$z**K#8K0088A;;>&HKegg,L--K&(:(:88A;;QWHXHX(Y(Y\c(cfn(nnK( ,$ '/x'7'7$$# '+$ A "> !;7777roriginal_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)rrZrrYc<g|]}|Sr0)r)rrUrr*s rrz3DPMSolverSDEScheduler.add_noise..!s*^^^qD33A7IJJ^^^rrr/)rrrrZtyper!is_floating_pointrr`rrr'flattenr& unsqueeze) r*rrrr step_indicesrK noisy_samplesrs ` @r add_noisezDPMSolverSDEScheduler.add_noises'7'>FVF\]]  " '5 0 0U5LY5W5W 0!%!2!23C3JRWR_!2!`!` ! %5%$=>>>L E$(#%BF(,1,0 *$,$, $,$, $,  $,  bj$u+&= >? $,$,$D>$,%SM$,$,$,$,$,$,N # # # #11133X3  X !!X!((3(((( u|+,  <,0-1 G"G" G"c5<'(G"&c] G"G"G"G"R    0 EL U\    ##X#! m8m8EL"*45m8u|+,m8elBJ./ m8  m8  m8 % &m8m8m8m8`,|<   B/////rri)rMrN)rQtypingrrrrnumpyrr!rconfiguration_utilsrr scheduling_utilsr r r rr@rgrir0rrrs5 //////////// AAAAAAAAXXXXXXXXXX++++++++6::::::::8!)4)4)4)4Xv/v/v/v/v/NKv/v/v/v/v/r