tiny_diff.schedulers.ddpm.LatentDDPMScheduler

class tiny_diff.schedulers.ddpm.LatentDDPMScheduler(num_train_steps: int = 1000, num_inference_steps: int | None = None, beta_start: float = 0.0001, beta_end: float = 0.02, clip_range: float | None = None, device: str | device | None = None)

Bases: DDPMScheduler

DDPM scheduler with original latent diffusion beta scheduler.

__init__(num_train_steps: int = 1000, num_inference_steps: int | None = None, beta_start: float = 0.0001, beta_end: float = 0.02, clip_range: float | None = None, device: str | device | None = None)

Methods

__init__([num_train_steps, ...])
add_noise(original_samples, noise, timesteps)	Adds noise in the forward diffusion process.
get_timesteps([n_steps, device])	Sets the discrete timesteps.
set_betas(beta_start, beta_end, n_steps[, ...])	Sets the beta schedule.
step(e, t, sample[, generator])	Predict noise at time t-1 given time t.
to(device)	Move to device.

Attributes

betas	Beta schedule.
clip_sample	Whether to clip a sample or not.
num_inference_steps	Diffusion steps to use in inference.
num_train_steps	Diffusion steps to use in train.