dmme.equations.ddim

reverse_process	Deterministic Denoising Process where \(\sigma_t = 0\) for all \(t\)
linear_tau	Linear sub-sequence \(\tau\)
quadratic_tau	Quadratic sub-sequence \(\tau\)

dmme.equations.ddim.reverse_process(x_t: Tensor, alpha_bar_t: Tensor, alpha_bar_t_minus_one: Tensor, noise_in_x_t: Tensor) → Normal

Deterministic Denoising Process where \(\sigma_t = 0\) for all \(t\)

Parameters:

• x_t – \(x_t\)

• alpha_bar_t – \(\bar\alpha_t\)

• alpha_bar_t_minus_one – \(\bar\alpha_{t-1}\) of shape \((N, 1, 1, *)\)

• noise_in_x_t – estimated noise in \(x_t\) predicted by a neural network

dmme.equations.ddim.linear_tau(timesteps: int, sub_timesteps: int) → Tensor

Linear sub-sequence \(\tau\)

Parameters:

• timesteps – total timesteps \(T\)

• sub_timesteps – sub-sequence length less than \(T\)

dmme.equations.ddim.quadratic_tau(timesteps: int, sub_timesteps: int) → Tensor

Quadratic sub-sequence \(\tau\)

Parameters:

• timesteps – total timesteps \(T\)

• sub_timesteps – sub-sequence length less than \(T\)