Struct PNDMScheduler 

pub struct PNDMScheduler {
    pub config: PNDMSchedulerConfig,
    /* private fields */
}

Pseudo numerical methods for diffusion models (PNDM) proposes using more advanced ODE integration techniques, namely Runge-Kutta method and a linear multi-step method.

Fields

config: PNDMSchedulerConfig

Implementations

impl PNDMScheduler

pub fn new(inference_steps: usize, config: PNDMSchedulerConfig) -> Self

pub fn timesteps(&self) -> &[usize]

pub fn scale_model_input(&self, sample: Tensor, _timestep: usize) -> Tensor

Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep.

pub fn step( &mut self, model_output: &Tensor, timestep: usize, sample: &Tensor, ) -> Tensor

pub fn add_noise( &self, original: &Tensor, noise: Tensor, timestep: usize, ) -> Tensor

pub fn init_noise_sigma(&self) -> f64