Official repository of ’Visual-RFT: Visual Reinforcement Fine-Tuning’
active 2025-03-04 → 2026-06-28 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 482 days from 2025-03-04 to 2026-06-28. Pushes: 26 total, peak 11 in a day. Pull requests: 3 total, peak 2 in a day. Issues: 124 total, peak 14 in a day. Comments: 253 total, peak 16 in a day. Stars: 921 total, peak 158 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| Liuziyu77 | 80 | 24 | 1 | 35 |
| lerogo | 16 | 0 | 0 | 11 |
| corkiyao | 12 | 0 | 0 | 9 |
| Justarrrrr | 11 | 0 | 0 | 9 |
| Qinyu-Allen-Zhao | 9 | 0 | 0 | 8 |
| JieFeng-cse | 7 | 0 | 0 | 5 |
| LilaKen | 7 | 0 | 0 | 7 |
| Chen-Song | 6 | 0 | 0 | 5 |
| flippy-bird | 5 | 0 | 0 | 3 |
| AliciaGe | 5 | 0 | 0 | 1 |
| JuzhengMiao | 5 | 0 | 0 | 4 |
| charlesuv | 5 | 0 | 0 | 4 |
| YiqunChen1999 | 5 | 0 | 0 | 5 |
| SunzeY | 5 | 2 | 1 | 2 |
| lsnls | 4 | 0 | 0 | 2 |
| FANGDUI0211 | 4 | 0 | 0 | 3 |
| pmf200103 | 4 | 0 | 0 | 3 |
| awakenlee180 | 4 | 0 | 0 | 4 |
| fake-warrior8 | 4 | 0 | 0 | 4 |
| yyyyysxz | 4 | 0 | 0 | 3 |
Recent activity
Latest issues, pull requests and releases
- Issue comment#250JuzhengMiao2025-12-19 10:57Difference between the GRPO implementation in this work and the original GRPO
- Issue comment#250childeyemax2025-12-19 08:09Difference between the GRPO implementation in this work and the original GRPO
- Issue#250JuzhengMiao2025-12-17 15:54Difference between the GRPO implementation in this work and the original GRPO
- Issue comment#245JuzhengMiao2025-12-17 03:29Why format rewads are always 0 using Qwen3-VL-8B-Instrcut
- Issue comment#191SII-WZ2025-12-08 07:23小白搭建环境报错求助
- Issue comment#238Re-dot-art2025-12-01 01:36lisa_grounding单卡顺畅，多卡就会卡在模型初始化
- Issue comment#245zhouxinwen-cpu2025-11-25 11:38Why format rewads are always 0 using Qwen3-VL-8B-Instrcut
- Issue comment#26zhouxinwen-cpu2025-11-25 11:09使用同样的参数复现，模型好像训崩了？
- Issue comment#232decimal22025-11-13 02:13Visual-RFT中训练的IoU Reward一直很低
- Issue comment#232Moyago2025-11-13 02:09Visual-RFT中训练的IoU Reward一直很低
- Issue#245ZhaoyangLi-12025-10-27 05:06Why format rewads are always 0 using Qwen3-VL-8B-Instrcut
- Issue comment#230lsnls2025-10-04 11:43qwenvl2.5环境报错
- Issue comment#133Ricardo0110112025-10-02 17:54No module named 'math_verify
- Issue#240rahulismyname2025-09-25 11:44SFT with LISA
- Issue#238GGbond-study2025-09-03 09:09lisa_grounding单卡顺畅，多卡就会卡在模型初始化
- Issue#237Cyril-ljx2025-08-21 11:44数据集的构建
- Issue comment#222psv6662025-08-15 08:29运行错误ModuleNotFoundError: No module named 'open_r1.trainer.grpo_trainer_visual_rft'
- Issue comment#233JaneM12222025-08-14 02:56细粒度分类复现中的loss和推理问题
- Pull request#236rahulismyname2025-08-13 19:00
- Issue#235Mountchicken2025-08-13 07:29Question about AP calculation in MLLM-based object detection without explicit scores
- Issue comment#223awakenlee1802025-08-12 04:07Visual-ARFT对于Qwen2-VL的适配
- Issue comment#93awakenlee1802025-08-12 04:02RuntimeError: Failed to import transformers.models.qwen2_vl.modeling_qwen2_vl because of the following error (look up to see its traceback)
- Issue comment#121awakenlee1802025-08-12 03:55IndexError: The shape of the mask [746] at index 0 does not match the shape of the indexed tensor [745] at index 0
- Issue comment#223awakenlee1802025-08-12 03:51Visual-ARFT对于Qwen2-VL的适配
- Issue comment#52JaneM12222025-08-11 06:18Missing"use_cache" leads to inconsistent model performance
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 921 stars here means stars gained during the window, not the repo's star count.