Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Multimodal RL training framework for diffusion & omni models
| Date | Stars |
|---|---|
| 2026-07-30 | 671 |
| 2026-08-06 | 671 |
| 2026-08-07 | 742 |
| 2026-08-15 | 794 |
| 2026-08-18 | 818 |
| 2026-08-19 | 821 |
| 2026-08-20 | 826 |
| 2026-08-21 | 836 |
| 2026-08-22 | 844 |
| 2026-08-23 | 849 |
| 2026-08-24 | 861 |
| 2026-08-25 | 867 |
| 2026-08-26 | 873 |
| 2026-08-27 | 880 |
| 2026-08-28 | 890 |
| 2026-08-29 | 896 |
| 2026-08-30 | 904 |
| 2026-08-31 | 912 |
| 2026-09-01 | 921 |
| 2026-09-02 | 933 |
| 2026-09-03 | 943 |
| 2026-09-04 | 949 |
| 2026-09-06 | 955 |
| 2026-09-07 | 959 |
| 2026-09-08 | 968 |
| 2026-09-09 | 982 |
| 2026-09-10 | 992 |
| 2026-09-11 | 1001 |
| 2026-09-12 | 1011 |
| 2026-09-13 | 1018 |
| 2026-09-14 | 1029 |
| 2026-09-15 | 1040 |
| 2026-09-16 | 1043 |
| 2026-09-17 | 1056 |
| 2026-09-18 | 1062 |
| 2026-09-19 | 1064 |
| 2026-09-20 | 1073 |
| 2026-09-21 | 1080 |
| 2026-09-22 | 1090 |
| 2026-09-23 | 1097 |
| 2026-09-24 | 1105 |
| 2026-09-25 | 1110 |
| 2026-09-26 | 1114 |
| 2026-09-27 | 1122 |
| 2026-09-28 | 1129 |
| 2026-09-29 | 1136 |
| 2026-09-30 | 1141 |
| 2026-10-01 | 1146 |
| 2026-10-02 | 1152 |
| 2026-10-03 | 1160 |
| 2026-10-04 | 1163 |
| 2026-10-05 | 1170 |
| 2026-10-06 | 1179 |
| 2026-10-07 | 1188 |
| 2026-10-08 | 1196 |
| 2026-10-09 | 1204 |
| 2026-10-10 | 1213 |
| 2026-10-11 | 1230 |
| 2026-10-12 | 1239 |
| 2026-10-13 | 1250 |
| 2026-10-14 | 1258 |
| 2026-10-15 | 1269 |
| 2026-10-16 | 1280 |
| 2026-10-17 | 1286 |
| 2026-10-18 | 1295 |
| 2026-10-19 | 1301 |
| 2026-10-20 | 1311 |
| 2026-10-21 | 1320 |
| 2026-10-22 | 1334 |
| 2026-10-23 | 1347 |
| 2026-10-24 | 1353 |
| 2026-10-25 | 1359 |
| 2026-10-26 | 1373 |
| 2026-10-27 | 1381 |
| 2026-10-28 | 1396 |
| 2026-10-29 | 1404 |
| 2026-10-30 | 1411 |
| 2026-10-31 | 1417 |
| 2026-11-01 | 1429 |
| 2026-11-02 | 1435 |
| 2026-11-03 | 1440 |
| 2026-11-04 | 1450 |
Today
+10 stars today
This week
+54 stars this week
This month
+280 stars this month
Momentum
129.0
growth rate 3.87%/day
<div align="center"> # VeRL-Omni ### Easy, fast, and stable RL training for diffusion and omni-modality models <a href="https://deepwiki.com/verl-project/verl-omni"><img src="https://devin.ai/assets/deepwiki-badge.png" alt="Ask DeepWiki.com" style="height:20px;"></a> [](https://verl-omni.readthedocs.io/en/latest/index.html) [](./LICENSE) <a href="docs/assets/WeChat.jpg"><img src="https://img.shields.io/badge/微信-green?logo=wechat"></a> <a href="https://join.slack.com/t/verl-project/shared_invite/zt-4bb32of1g-c~szeecZ2fufGeSadsIefw"><img src="https://img.shields.io/badge/Slack-verl-blueviolet?logo=slack"></a> <a href="https://drive.google.com/file/d/1EGVFZdvRgSBymMlZa62FObbzEhMoLA0F/view?usp=sharing"><img src="https://img.shields.io/badge/Slides-verl--omni-orange?logo=google-drive&logoColor=white"></a> </div> `VeRL-Omni` is a general RL training framework focused on multimodal generative models, built on top of [`verl`](https://github.com/verl-project/verl). It originated from the multi-modal generation RL effort in `verl`, and now has a dedicated home so it can evolve in a more focused way. ## News 🔥 - **[2026-09]** The [v1 trainer](docs/start/diffusion_v1.md) (TransferQueue + ReplayBuffer) is now the default for diffusion models; the legacy v0 trainer is deprecated, while offline DPO remains on v0 by design. - **[2026-08]** **[DiffusionOPD](docs/algo/diffusion_opd.md)** (on-policy distillation, including multi-teacher MOPD) is now supported. - **[2026-08]** 🔥 **MiniMax-H3** now supports T2VA, FL2VA, and Ref2VA with both [FlowGRPO](examples/flowgrpo_trainer/minimax_h3/README.md) and [DiffusionNFT](examples/diffusionnft_trainer/minimax_h3/README.md). - **[2026-08]** 🎉 We have released [v0.2.0](https://github.com/verl-project/verl-omni/releases/tag/v0.2.0) for faster diffusion rl and more stable Qwen3-Omni multimodal training. Blog: [VeRL-Omni v0.2.0](https://verl-project.github.io/posts/2026-08-17-verl-omni-v0-2-0/) - **[2026-08]** [LTX2.3](examples/flowgrpo_trainer/ltx2/README.md) text-to-video+audio model is now supported with FlowGRPO. - **[2026-07]** Team-proposed algorithm [FlowGRPO with DiNa-LRM](https://verl-omni.readthedocs.io/en/latest/examples/flowgrpo_trainer_sd35_drm.html) is available. Training skips VAE decoding by scoring clean diffusion latents directly for faster and more resource-efficient model alignment. - **[2026-07]** VeRL-Omni is presented in QingKe AI, vLLM community, and verl x Ascend Beijing meetup. [Slides](https://drive.google.com/file/d/1RJBZZ2k6exxciFghU1FNUgds6s1w7Bpf/view?usp=sharing) are shared. - **[2026-06]** [Qwen3-Omni GSPO Trainer](examples/gspo_trainer) is available! [Flow-DPPO](https://verl-omni.readthedocs.io/en/latest/algo/flowdppo.html) is integrated. vLLM-Omni rollout backend is upgraded to v0.22 for higher throughput, with default actor attn backend switched to FA3. - **[2026-06]** [DiffusionNFT](https://verl-omni.readthedocs.io/en/latest/algo/diffusionnft.html) and [Diffusion DPO](https://verl-omni.readthedocs.io/en/latest/algo/diffusion_dpo.html) are integrated with verified recipes on Qwen-Image/SD3.5. [Wan2.2](examples/dancegrpo_trainer/README.md) is now supported for video generation tasks. ## Why `VeRL-Omni` Multimodal generative RL training differs from text-only LLM RL not only in model structure, but also in I/O patterns, compute characteristics, and runtime bottlenecks. As this space grows, it deserves a dedicated training repository that can evolve quickly around its own constraints. ### Scope `VeRL-Omni` targets RL post-training for three families of generative models: 1. **Diffusion generative models** for image, video, and audio — e.g., Qwen-Image, Wan2.2. 2. **Unified multimodal understanding + generation models** — e.g., BAGEL, HunyuanImage-3.0. 3. **Omni-modality models** that jointly handle text, image, audio, and video
Excerpt of 10,963 characters
Read on GitHub
61
Samit · HKUST · Hong Kong
61
31
21
10
9
9
7
4
3
3
3
3
汪志鹏 · Rakuten Asia Pte Ltd · Singapore
2
2
2
1
1
1
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:ea8a30838b2c3863, llm:Description and README: 'Multimodal RL training framework for diffusion & omni models', 'general RL training framework focused on multimodal generative models', topics include reinforcement-learning, rlhf, flow-matching, diffusion-models, multimodal, vllm, qwen.
matched fp:ea8a30838b2c3863, llm:Description and README: 'Multimodal RL training framework for diffusion & omni models', 'general RL training framework focused on multimodal generative models', topics include reinforcement-learning, rlhf, flow-matching, diffusion-models, multimodal, vllm, qwen.
matched fp:ea8a30838b2c3863, llm:Description and README: 'Multimodal RL training framework for diffusion & omni models', 'general RL training framework focused on multimodal generative models', topics include reinforcement-learning, rlhf, flow-matching, diffusion-models, multimodal, vllm, qwen.