def model_and_diffusion_defaults()

in improved_diffusion/script_util.py [0:0]

• 22 lines of code
• 1 McCabe index (conditional complexity)

def model_and_diffusion_defaults():
    """
    Defaults for image training.
    """
    return dict(
        image_size=64,
        num_channels=128,
        num_res_blocks=2,
        num_heads=4,
        num_heads_upsample=-1,
        attention_resolutions="16,8",
        dropout=0.0,
        learn_sigma=False,
        sigma_small=False,
        class_cond=False,
        diffusion_steps=1000,
        noise_schedule="linear",
        timestep_respacing="",
        use_kl=False,
        predict_xstart=False,
        rescale_timesteps=True,
        rescale_learned_sigmas=True,
        use_checkpoint=False,
        use_scale_shift_norm=True,
    )