data

DDPMConfig

extendsDiffusionModelConfig

DDPMConfig(sample_size: int | tuple[int, int] = 32, in_channels: int = 3, out_channels: int = 3, act_fn: GenerativeActivation = 'silu', num_train_timesteps: int = 1000, beta_start: float = 0.0001, beta_end: float = 0.02, beta_schedule: BetaSchedule = 'linear', prediction_type: Literal['epsilon', 'sample', 'v_prediction'] = 'epsilon', base_channels: int = 128, channel_mult: tuple[int, ...] = (lambda: (1, 2, 2, 2))(), num_res_blocks: int = 2, attention_resolutions: tuple[int, ...] = (lambda: (16,))(), num_heads: int = 1, dropout: float = 0.1, resnet_groups: int = 32, learn_sigma: bool = False, clip_denoised: bool = True)

source edit

Configuration for every DDPM variant.

Parameters

base_channelsint= 128

Channel width of the first U-Net stage C_0. All other stages are integer multiples (see channel_mult). Paper default: 128.

channel_multtuple[int, ...]= (lambda: (1, 2, 2, 2))()

Per-stage channel multiplier. (1, 2, 2, 2) for CIFAR (Table 9), (1, 1, 2, 2, 4, 4) for LSUN/CelebA-HQ. Length = number of downsampling stages; spatial resolution divides by 2 ** (L - 1) (no downsample after the final stage).

num_res_blocksint= 2

ResBlocks per stage (paper: 2).

attention_resolutionstuple[int, ...]= (lambda: (16,))()

Spatial sizes at which self-attention is applied. Ho 2020 uses (16,) — attention only at the 16×16 feature scale.

num_headsint= 1

Attention head count (paper: 1 for CIFAR, can scale).

dropoutfloat= 0.1

Dropout inside ResBlocks. Paper: 0.1 (CIFAR), 0.0 (CelebA / LSUN).

resnet_groupsint= 32

GroupNorm group count inside ResBlocks (paper default: 32). Must divide every per-stage channel count.

learn_sigmabool= False

Improved-DDPM (Nichol 2021) — when True, the network predicts the variance in addition to the mean. Output channels become 2 * in_channels.

clip_denoisedbool= True

Clip the predicted x_0 to [-1, 1] during sampling (default True).

Used by 4

Constructors

dunder

init

→None

__init__(sample_size: int | tuple[int, int] = 32, in_channels: int = 3, out_channels: int = 3, act_fn: GenerativeActivation = 'silu', num_train_timesteps: int = 1000, beta_start: float = 0.0001, beta_end: float = 0.02, beta_schedule: BetaSchedule = 'linear', prediction_type: Literal['epsilon', 'sample', 'v_prediction'] = 'epsilon', base_channels: int = 128, channel_mult: tuple[int, ...] = (lambda: (1, 2, 2, 2))(), num_res_blocks: int = 2, attention_resolutions: tuple[int, ...] = (lambda: (16,))(), num_heads: int = 1, dropout: float = 0.1, resnet_groups: int = 32, learn_sigma: bool = False, clip_denoised: bool = True)

source edit

Properties

prop

out_channels_effective

→int

out_channels_effective: int

source edit

How many channels the U-Net actually emits (doubled when learn_sigma=True).

DDPMConfig(sample_size: int | tuple[int, int] = 32, in_channels: int = 3, out_channels: int = 3, act_fn: GenerativeActivation = 'silu', num_train_timesteps: int = 1000, beta_start: float = 0.0001, beta_end: float = 0.02, beta_schedule: BetaSchedule = 'linear', prediction_type: Literal['epsilon', 'sample', 'v_prediction'] = 'epsilon', base_channels: int = 128, channel_mult: tuple[int, ...] = (lambda: (1, 2, 2, 2))(), num_res_blocks: int = 2, attention_resolutions: tuple[int, ...] = (lambda: (16,))(), num_heads: int = 1, dropout: float = 0.1, resnet_groups: int = 32, learn_sigma: bool = False, clip_denoised: bool = True)

__init__(sample_size: int | tuple[int, int] = 32, in_channels: int = 3, out_channels: int = 3, act_fn: GenerativeActivation = 'silu', num_train_timesteps: int = 1000, beta_start: float = 0.0001, beta_end: float = 0.02, beta_schedule: BetaSchedule = 'linear', prediction_type: Literal['epsilon', 'sample', 'v_prediction'] = 'epsilon', base_channels: int = 128, channel_mult: tuple[int, ...] = (lambda: (1, 2, 2, 2))(), num_res_blocks: int = 2, attention_resolutions: tuple[int, ...] = (lambda: (16,))(), num_heads: int = 1, dropout: float = 0.1, resnet_groups: int = 32, learn_sigma: bool = False, clip_denoised: bool = True)