flow_matching_cifar

→FlowMatchingModel

flow_matching_cifar(pretrained: bool = False, overrides: object = {})

source edit

Flow Matching velocity field for the CIFAR-10 setup.

Trained by regressing onto the optimal-transport conditional field — no ODE is solved during training, which is the method's whole point. The solver appears only when sampling or scoring likelihood.

Model Size

Parameters

pretrainedbool= False

Reserved for future weight registration; currently a no-op.

**overridesobject= {}

Optional FlowMatchingConfig field overrides forwarded into the underlying config.

Returns

FlowMatchingModel

Velocity field configured for 32 x 32 RGB.

Notes

Reference: Lipman, Chen, Ben-Hamu, Nickel, and Le, "Flow Matching for Generative Modeling", ICLR, 2023 (arXiv:2210.02747), Table 1. Reported for the optimal-transport path: 6.35 FID / 2.99 bits per dimension.

Examples

>>> import lucid
>>> from lucid.models.generative.flow_matching import flow_matching_cifar
>>> model = flow_matching_cifar(base_channels=16, channel_mult=(1, 2),
...                             num_res_blocks=1, sample_size=8,
...                             attention_resolutions=(),
...                             resnet_groups=8).eval()
>>> loss, _, _ = model.flow_matching_loss(lucid.randn((2, 3, 8, 8)))
>>> model.nfe                      # training solves nothing
0

Used by 1

lucid.models.generative.flow_matching