# HeunDiscreteScheduler ## HeunDiscreteScheduler The Heun scheduler (Algorithm 1) is from the [Elucidating the Design Space of Diffusion-Based Generative Models](https://huggingface.co/papers/2206.00364) paper by Karras et al. The scheduler is ported from the [k-diffusion](https://github.com/crowsonkb/k-diffusion) library and created by [Katherine Crowson](https://github.com/crowsonkb/). ### HeunDiscreteScheduler #### class diffusers.HeunDiscreteScheduler [\](https://github.com/huggingface/diffusers/blob/v0.21.0/src/diffusers/schedulers/scheduling_heun_discrete.py#L71) ( num\_train\_timesteps: int = 1000beta\_start: float = 0.00085beta\_end: float = 0.012beta\_schedule: str = 'linear'trained\_betas: typing.Union\[numpy.ndarray, typing.List\[float], NoneType] = Noneprediction\_type: str = 'epsilon'use\_karras\_sigmas: typing.Optional\[bool] = Falseclip\_sample: typing.Optional\[bool] = Falseclip\_sample\_range: float = 1.0timestep\_spacing: str = 'linspace'steps\_offset: int = 0 ) Parameters * **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`. * **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). * **clip\_sample** (`bool`, defaults to `True`) — Clip the predicted sample for numerical stability. * **clip\_sample\_range** (`float`, defaults to 1.0) — The maximum magnitude for sample clipping. Valid only when `clip_sample=True`. * **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}. * **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. You can use a combination of `offset=1` and `set_alpha_to_one=False` to make the last step use step 0 for the previous alpha product like in Stable Diffusion. Scheduler with Heun steps for discrete beta schedules. This model inherits from [SchedulerMixin](https://huggingface.co/docs/diffusers/v0.21.0/en/api/schedulers/overview#diffusers.SchedulerMixin) and [ConfigMixin](https://huggingface.co/docs/diffusers/v0.21.0/en/api/configuration#diffusers.ConfigMixin). Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving. **scale\_model\_input** [\](https://github.com/huggingface/diffusers/blob/v0.21.0/src/diffusers/schedulers/scheduling_heun_discrete.py#L187) ( sample: FloatTensortimestep: typing.Union\[float, torch.FloatTensor] ) → `torch.FloatTensor` Parameters * **sample** (`torch.FloatTensor`) — The input sample. * **timestep** (`int`, *optional*) — The current timestep in the diffusion chain. Returns `torch.FloatTensor` A scaled input sample. Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. **set\_timesteps** [\](https://github.com/huggingface/diffusers/blob/v0.21.0/src/diffusers/schedulers/scheduling_heun_discrete.py#L213) ( num\_inference\_steps: intdevice: typing.Union\[str, torch.device] = Nonenum\_train\_timesteps: typing.Optional\[int] = None ) Parameters * **num\_inference\_steps** (`int`) — The number of diffusion steps used when generating samples with a pre-trained model. * **device** (`str` or `torch.device`, *optional*) — The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. Sets the discrete timesteps used for the diffusion chain (to be run before inference). **step** [\](https://github.com/huggingface/diffusers/blob/v0.21.0/src/diffusers/schedulers/scheduling_heun_discrete.py#L340) ( model\_output: typing.Union\[torch.FloatTensor, numpy.ndarray]timestep: typing.Union\[float, torch.FloatTensor]sample: typing.Union\[torch.FloatTensor, numpy.ndarray]return\_dict: bool = True ) → [SchedulerOutput](https://huggingface.co/docs/diffusers/v0.21.0/en/api/schedulers/singlestep_dpm_solver#diffusers.schedulers.scheduling_utils.SchedulerOutput) or `tuple` Parameters * **model\_output** (`torch.FloatTensor`) — The direct output from learned diffusion model. * **timestep** (`float`) — The current discrete timestep in the diffusion chain. * **sample** (`torch.FloatTensor`) — A current instance of a sample created by the diffusion process. * **return\_dict** (`bool`) — Whether or not to return a [SchedulerOutput](https://huggingface.co/docs/diffusers/v0.21.0/en/api/schedulers/singlestep_dpm_solver#diffusers.schedulers.scheduling_utils.SchedulerOutput) or tuple. Returns [SchedulerOutput](https://huggingface.co/docs/diffusers/v0.21.0/en/api/schedulers/singlestep_dpm_solver#diffusers.schedulers.scheduling_utils.SchedulerOutput) or `tuple` If return\_dict is `True`, [SchedulerOutput](https://huggingface.co/docs/diffusers/v0.21.0/en/api/schedulers/singlestep_dpm_solver#diffusers.schedulers.scheduling_utils.SchedulerOutput) is returned, otherwise a tuple is returned where the first element is the sample tensor. Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned model outputs (most often the predicted noise). ### SchedulerOutput #### class diffusers.schedulers.scheduling\_utils.SchedulerOutput [\](https://github.com/huggingface/diffusers/blob/v0.21.0/src/diffusers/schedulers/scheduling_utils.py#L50) ( prev\_sample: FloatTensor ) Parameters * **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. Base class for the output of a scheduler’s `step` function.