o pi"@sddlZddlmZddlmZmZmZmZddlZddl m Z m Z ddl m Z ddlmZdd lmZeGd d d e Z dddZGdddee ZdS)N) dataclass)ListOptionalTupleUnion) ConfigMixinregister_to_config) BaseOutput) randn_tensor)SchedulerMixinc@seZdZUdZejed<dS)DDPMWuerstchenSchedulerOutputaA 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. prev_sampleN)__name__ __module__ __qualname____doc__torchTensor__annotations__rrm/home/ubuntu/SoloSpeech/.venv/lib/python3.10/site-packages/diffusers/schedulers/scheduling_ddpm_wuerstchen.pyrs  r+?cosinecCs|dkr dd}n|dkrdd}ntd|g}t|D]}||}|d|}|td|||||qtj|tjdS) 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 rcSs t|ddtjddS)NMb?gT㥛 ?r)mathcospitrrr alpha_bar_fnEs z)betas_for_alpha_bar..alpha_bar_fnexpcSst|dS)Ng()rr"rrrrr!Jsz"Unsupported alpha_transform_type: r dtype) ValueErrorrangeappendminrtensorfloat32)num_diffusion_timestepsmax_betaalpha_transform_typer!betasit1t2rrrbetas_for_alpha_bar,s    "r2c @seZdZdZe  d"dedefddZdd Zd#d ej d e e d ej fddZ d$de de e e deeejffddZ d%dej d e d ej ded eeeff ddZdej dej dej d ej fddZddZd d!Zd S)&DDPMWuerstchenSchedulera Denoising diffusion probabilistic models (DDPMs) explores the connections between denoising score matching and Langevin dynamics sampling. [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__` function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`. [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and [`~SchedulerMixin.from_pretrained`] functions. For more details, see the original paper: https://arxiv.org/abs/2006.11239 Args: scaler (`float`): .... s (`float`): .... ?rscalerscCsD||_t|g|_t|jd|jtjdd|_d|_dS)Nr ?rr4)r5rr)r6rr_init_alpha_cumprodinit_noise_sigma)selfr5r6rrr__init__is& z DDPMWuerstchenScheduler.__init__cCs~|jdkrdd||j}n |jdkr||j}t||j|d|j|tjdd|j|}|ddS)Nr r7rg-C6?gH.?)r5rrr6torr8clamp)r:r device alpha_cumprodrrr_alpha_cumprodvs   (  z&DDPMWuerstchenScheduler._alpha_cumprodNsampletimestepreturncCs|S)aP Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. Args: sample (`torch.Tensor`): input sample timestep (`int`, optional): current timestep Returns: `torch.Tensor`: scaled input sample r)r:rArBrrrscale_model_inputs z)DDPMWuerstchenScheduler.scale_model_inputnum_inference_steps timestepsr>cCsD|durtjdd|d|d}t|tjst||}||_dS)a Sets the discrete timesteps used for the diffusion chain. Supporting function to be run before inference. Args: num_inference_steps (`Dict[float, int]`): the number of diffusion steps used when generating samples with a pre-trained model. If passed, then `timesteps` must be `None`. device (`str` or `torch.device`, optional): the device to which the timesteps are moved to. {2 / 3: 20, 0.0: 10} Nr4gr r>)rlinspace isinstancerr<rF)r:rErFr>rrr set_timestepss   z%DDPMWuerstchenScheduler.set_timestepsT model_output return_dictcCsD|j}|j}|}||} |||j|dgdd|jddDR} || |j| dgdd|jddDR} | | } d| |d| |d| } t| j||j|jd}d| d| d| |}| || dk j| dgd d|jddDR}|s| |fSt | |d S) a~ Predict the sample at the previous timestep by reversing the SDE. Core function to propagate the diffusion process from the learned model outputs (most often the predicted noise). Args: model_output (`torch.Tensor`): direct output from learned diffusion model. timestep (`int`): current discrete timestep in the diffusion chain. sample (`torch.Tensor`): current instance of sample being created by diffusion process. generator: random number generator. return_dict (`bool`): option for returning tuple rather than DDPMWuerstchenSchedulerOutput class Returns: [`DDPMWuerstchenSchedulerOutput`] or `tuple`: [`DDPMWuerstchenSchedulerOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is the sample tensor. rcSg|]}dqSr r.0_rrr z0DDPMWuerstchenScheduler.step..r NcSrMrNrrOrrrrRrSr4) generatorr>r$cSrMrNrrOrrrrRrS)r) r$r>previous_timestepr@viewsizeshapesqrtr floatr<r)r:rKrBrArTrLr$r>r prev_tr?alpha_cumprod_prevalphamu std_noisestdpredrrrsteps 44( < zDDPMWuerstchenScheduler.steporiginal_samplesnoisecCsj|j}|j}|j||dj|dgdd|jddDR}||d||}|j|dS)NrGrcSrMrNrrOrrrrRrSz5DDPMWuerstchenScheduler.add_noise..r r#)r>r$r@rVrWrXrYr<)r:rcrdrFr>r$r? noisy_samplesrrr add_noises z!DDPMWuerstchenScheduler.add_noisecCs|jjSN)confignum_train_timesteps)r:rrr__len__szDDPMWuerstchenScheduler.__len__cCs<|j|d}|j|dd|jd}|S)Nrr )rFabsargminitemexpandrX)r:rBindexr[rrrrUsz)DDPMWuerstchenScheduler.previous_timestep)r4rrg)NNN)NT)rrrrr rZr;r@rrrintrDrrstrr>rJboolrrrbrfrjrUrrrrr3Xs\       .  r3)rr)r dataclassesrtypingrrrrrconfiguration_utilsrr utilsr utils.torch_utilsr scheduling_utilsr rr2r3rrrrs     ,