# Tencent-Hunyuan/MixGRPO

[ECCV 2026] MixGRPO: Unlocking Flow-based GRPO Efficiency with Mixed ODE-SDE

Repository: https://github.com/Tencent-Hunyuan/MixGRPO
Canonical: https://ross.abutalabs.com/products/mixgrpo
Homepage: https://tulvgengenr.github.io/MixGRPO-Project-Page/
Language: Python
License: NOASSERTION
License Family: other
Topics: diffusion, grpo, reinforcement-learning
Last push: 2026-07-01T10:49:26+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 90, release rhythm 35, longevity 28
- inputs: {"age_days": 400, "days_push": 63, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1177, forks 51 (observed 2026-08-28T04:03:52.724118+00:00)

## What it is
MixGRPO is a research framework from Tencent Hunyuan implementing a mixed ODE-SDE GRPO algorithm for efficient reinforcement learning fine-tuning of flow-based diffusion image generation models. It includes training code, a FLUX.1 Dev fine-tuned checkpoint, and a faster MixGRPO-Flash variant using a sliding window mechanism.

## Use cases
- fine-tune diffusion models with GRPO for human preference alignment
- speed up flow-based RL post-training of image generators
- reproduce MixGRPO results from the ECCV 2026 paper
- train SD3.5 or FLUX LoRA with reward models like HPSv2
- compare Flow-GRPO and MixGRPO training efficiency

## When to choose
- you need RL-based preference alignment for flow matching image models
- you want faster GRPO training with fewer denoising steps optimized
- you want the official implementation and checkpoints of the MixGRPO paper

## When to avoid
- you need a production-ready, well-licensed training framework
- you are not working with diffusion or flow-based generative models
- you need stable long-term support rather than research code

## Facets
- artifact type: library
- maturity: active
- function: reinforcement-learning, machine-learning, llm-training, image-processing
- domain: machine-learning, deep-learning, image-processing, artificial-intelligence
- platform: python
- tags: diffusion-models, grpo, flow-matching, rlhf, image-generation, research-code, eccv-2026, gpu, linux

## Member repositories
- Tencent-Hunyuan/MixGRPO (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:52.724118+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:26:10.990589+00:00, confidence not recorded.
  - readme: https://github.com/Tencent-Hunyuan/MixGRPO (fetched 2026-08-28T04:03:52.724118+00:00, sha 254902b71ee1)
  - homepage: https://tulvgengenr.github.io/MixGRPO-Project-Page/ (fetched 2026-08-29T12:33:02.554378+00:00, sha 98e4f75a90fc)
- Data as of 2026-08-30T08:39:29.467469+00:00.