ddpm_cifar_gen

→DDPMForImageGeneration

ddpm_cifar_gen(pretrained: bool | str = False, weights: DDPMCifarWeights | None = None, overrides: object = {})

source edit

Construct a DDPM CIFAR-10 model with training loss and .generate().

Same trunk as ddpm_cifar (sample size 32x32, 4-stage U-Net), wrapped with the Ho 2020 simplified training objective and DiffusionMixin.generate for ancestral sampling.

Model Size

Parameters

pretrainedbool or str= False

Weight selector. False → random init; True → the official google/ddpm-cifar10-32 checkpoint (DDPMCifarWeights.CIFAR10), giving an inference-ready sampler — model.generate(...) draws CIFAR-10-like images.

weights(DDPMCifarWeights, optional, keyword - only)= None

Explicit weights enum member; takes precedence over pretrained.

**overridesobject= {}

Optional DDPMConfig field overrides forwarded into the underlying config.

Returns

DDPMForImageGeneration

CIFAR-10 DDPM wrapped with the noise-prediction loss head (pretrained when requested).

Notes

Reference: Ho, Jain, and Abbeel, "Denoising Diffusion Probabilistic Models", NeurIPS, 2020 (arXiv:2006.11239), Appendix B.1.

Examples

>>> import lucid
>>> from lucid.models.generative.ddpm import ddpm_cifar_gen
>>> model = ddpm_cifar_gen().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)
(1, 3, 32, 32)

Used by 1

lucid.models.generative.ddpm