Score-Based Generative Modeling through SDEs

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1844 stars · 232 forks observed · 2026-08-28

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

Official PyTorch implementation of the ICLR 2021 paper 'Score-Based Generative Modeling through Stochastic Differential Equations', unifying score-based and diffusion generative models via SDEs. It supports training NCSN++, DDPM++, and earlier score-based models, sampling, likelihood evaluation, and conditional generation tasks like inpainting, colorization, and class-conditional synthesis.

Use cases

• train a score-based diffusion model on images
• generate CIFAR-10 or CelebA-HQ samples with an SDE model
• compute exact likelihoods with a neural ODE
• solve inverse problems like image inpainting and colorization
• reproduce NCSN, NCSNv2, DDPM, and DDPM++ baselines
• experiment with custom SDEs, predictors, and correctors

When to choose

• you need a reference implementation of score-based SDE generative models for research
• you want to train or evaluate diffusion/score models with predictor-corrector samplers
• you need exact likelihood computation or controllable generation from a pretrained score model

When to avoid

• you want a production-ready, actively maintained diffusion library - use Hugging Face Diffusers instead
• you need fast, optimized sampling for deployment rather than research experimentation
• you work outside image generation and cannot adapt the model architectures

Facets

library · maturity maintenance

machine-learning deep-learning image-processing data-generation deep-learning machine-learning image-processing artificial-intelligence python diffusion-models score-based-generative-modeling stochastic-differential-equations score-matching pytorch research-code iclr-2021 inverse-problems controllable-generation gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="yang-song/score_sde")