The Linear Schedule suggests that any number of time-steps are supported. But this is not true, since the Guided Diffusion asserts that all betas are in the interval (0,1). So any num_diffusion_timesteps<200 will yield an assertion error.
To reproduce, train any model with the flag diffusion_steps < 200.
The Linear Schedule suggests that any number of time-steps are supported. But this is not true, since the Guided Diffusion asserts that all betas are in the interval (0,1). So any
num_diffusion_timesteps<200will yield an assertion error.To reproduce, train any model with the flag
diffusion_steps < 200.