haidog-yaqub/MeanFlow

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1196 stars · 70 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

An unofficial PyTorch implementation of MeanFlow and iMF, one-step generative modeling methods based on flow matching. It provides config-driven training scripts for MNIST, CIFAR-10, and ImageNet latent generation with multi-GPU support via Accelerate.

Use cases

• train a one-step generative image model
• reproduce MeanFlow paper results in PyTorch
• generate MNIST or CIFAR-10 images with a single sampling step
• experiment with flow matching and CFG distillation
• train latent ImageNet generation models on multiple GPUs

When to choose

• you want a PyTorch (rather than JAX) implementation of MeanFlow or iMF
• you need one-step generative modeling with classifier-free guidance
• you want ready-made configs for MNIST, CIFAR-10, or ImageNet latent training

When to avoid

• you need the official, fully supported implementation
• you require Triton JVP kernels or Flash Attention within JVP, which are not yet supported
• you need a production inference library rather than research training code

Facets

library · maturity active

machine-learning deep-learning image-processing deep-learning machine-learning image-processing python diffusion-models flow-matching generative-modeling pytorch one-step-generation research-code gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="haidog-yaqub/MeanFlow")