ddim_scheduler_create

Create a DDIM Scheduler

Description

Creates a Denoising Diffusion Implicit Models (DDIM) scheduler for use with diffusion models. DDIM schedulers provide a deterministic sampling process that offers faster inference compared to DDPM while maintaining high quality outputs.

Usage

ddim_scheduler_create(num_train_timesteps = 1000, num_inference_steps = 50,
                      eta = 0,
                      beta_schedule = c("linear", "scaled_linear", "cosine"),
                      beta_start = 0.00085, beta_end = 0.012,
                      rescale_betas_zero_snr = FALSE,
                      dtype = torch::torch_float32(),
                      device = c(torch::torch_device("cpu"), torch::torch_device("cuda")))

Arguments

num_train_timesteps: Integer. The number of diffusion steps used to train the model. Default: 1000
num_inference_steps: Integer. The number of diffusion steps used for inference. Fewer steps typically means faster inference at the cost of sample quality. Default: 50
eta: Numeric. Controls the amount of stochasticity. When eta=0, the sampling process is deterministic. When eta=1, the sampling process is equivalent to DDPM. Default: 0
beta_schedule: Character. The beta schedule to use. Options are:
“linear”: Linear beta schedule from beta_start to beta_end
“scaled_linear”: Scaled linear schedule, generally gives better results
“cosine”: Cosine schedule that approaches zero smoothly Default: “linear”
beta_start: Numeric. The starting value for the beta schedule. Default: 0.00085
beta_end: Numeric. The final value for the beta schedule. Default: 0.012
rescale_betas_zero_snr: Logical. If TRUE, rescales the beta values
dtype: The data type to use for computations. Default is torch_float32(). Options are torch_float16() and torch_float32().
device: The device to use for computations. Options are torch_device(“cpu”), torch_device(“cuda”).

Details

DDIM (Denoising Diffusion Implicit Models) was introduced by Song et al. (2020) as an extension to DDPM (Denoising Diffusion Probabilistic Models). It offers a deterministic sampling process and allows for controlling the number of inference steps independently from the training process.

The scheduler contains the noise schedule and methods for computing alpha, beta, and other parameters used in the diffusion process.

Value

A DDIM scheduler object that can be used with diffusion models to generate samples.

References

Song, J., Meng, C., & Ermon, S. (2020). “Denoising Diffusion Implicit Models.” https://arxiv.org/abs/2010.02502

Examples

# Create a DDIM scheduler with custom parameters
scheduler <- ddim_scheduler_create(
  num_train_timesteps = 1000,
  num_inference_steps = 30,
  eta = 0.5,
  beta_schedule = "scaled_linear"
)

Packages