class

DDPMModel

extendsPretrainedModelDiffusionMixin

DDPMModel(config: DDPMConfig)

source edit

Bare DDPM U-Net trunk with the sampling-capable DiffusionMixin.

Returns a DiffusionModelOutput whose sample field is the raw network output — interpreted as predicted noise $\epsilon$ , clean signal $x_0$ , or velocity $v$ depending on config.prediction_type. When config.learn_sigma is True the output channels are 2 * in_channels: the first half is the mean prediction, the second half the (raw) variance prediction; downstream code splits the two if needed.

Parameters

configDDPMConfig

Hyperparameters governing the U-Net topology and the diffusion noise schedule (num_train_timesteps, beta_start, beta_end, beta_schedule, prediction_type).

Attributes

unetDDPMUNet

Underlying U-Net denoiser (see DDPMUNet).

config_classtype[DDPMConfig]

Registry pointer for matching-config instantiation.

base_model_prefixstr

State-dict prefix ("ddpm") under which the trunk is nested in task-head variants.

Notes

Reference: Ho, Jain, and Abbeel, "Denoising Diffusion Probabilistic Models", NeurIPS, 2020 (arXiv:2006.11239).

The forward Markov chain has the closed-form marginal

q(x_t \mid x_0) = \mathcal{N}\!\big( x_t;\; \sqrt{\bar{\alpha}_t}\, x_0,\; (1 - \bar{\alpha}_t)\, \mathbf{I} \big),

with $\bar{\alpha}_t = \prod_{s=1}^{t} (1 - \beta_s)$ . The U-Net is trained to predict the injected noise so reverse sampling iterates $t = T, T-1, \ldots, 1$ via the standard ancestral Gaussian update.

Examples

>>> import lucid
>>> from lucid.models.generative.ddpm import DDPMConfig, DDPMModel
>>> cfg = DDPMConfig(sample_size=32, base_channels=32,
...                  channel_mult=(1, 2), num_res_blocks=1,
...                  resnet_groups=16, num_train_timesteps=100)
>>> model = DDPMModel(cfg).eval()
>>> x_t = lucid.randn((1, 3, 32, 32))
>>> t = lucid.tensor([42]).long()
>>> out = model(x_t, t)
>>> out.sample.shape   # (1, 3, 32, 32) — predicted noise
(1, 3, 32, 32)

Used by 2

Constructors

dunder

init

→None

__init__(config: DDPMConfig)

source edit

Instance methods

forward

→DiffusionModelOutput

forward(sample: Tensor, timestep: Tensor)

source edit

Predict noise (or [noise, σ²]) at the given timestep.

Parameters

sample

(B, in_channels, H, W) noisy image x_t.

timestep

(B,) or () int timestep tensor. Scalar tensors are broadcast.

Returns

DiffusionModelOutput

(B, out_channels_effective, H, W) raw network output.

>>> import lucid >>> from lucid.models.generative.ddpm import DDPMConfig, DDPMModel >>> cfg = DDPMConfig(sample_size=32, base_channels=32, ... channel_mult=(1, 2), num_res_blocks=1, ... resnet_groups=16, num_train_timesteps=100) >>> model = DDPMModel(cfg).eval() >>> x_t = lucid.randn((1, 3, 32, 32)) >>> t = lucid.tensor([42]).long() >>> out = model(x_t, t) >>> out.sample.shape # (1, 3, 32, 32) — predicted noise (1, 3, 32, 32)