v0.2.0

What's Changed

• add test_od_r1 by @KingSan666888 in #157
• Develop/v0.2.0 by @SZhanZ in #165
• add math model in README by @SZhanZ in #166
• add features in README by @SZhanZ in #170
• Sync the blog content. by @SZhanZ in #176

Full Changelog: v0.1.0...v0.2.0

v0.1.0

What's Changed

• Ruox/main jsonl dataloader by @xrc10 in #14
• docs: update README.md by @eltociear in #27
• fix model torch_dtype setting by @zhangqianqianhzlh in #49
• custom reward by @zhangqianqianhzlh in #55
• multi-node GPRO recipe by @xrc10 in #59
• add epsilon clipping for GRPO by @xrc10 in #61
• formats are not unique by @Amos1109 in #65
• Add num_iterations from original GRPO algorithm by @xrc10 in #78
• Update grpo_trainer.py by @davidluciolu in #58
• add yes_no_reward function by @KingSan666888 in #88
• fix default reward method by @zhangqianqianhzlh in #87
• fix multi match case by @zhangqianqianhzlh in #94
• convert other types of answers by @xrc10 in #95
• fix data loader and batching by @xrc10 in #98
• fix: pin transformers to v4.49.0 to resolve model loading issues by @chaoyuhao in #105
• llm reward by @Amos1109 in #115
• math_reward by @Amos1109 in #127
• support language training data by @zhangqianqianhzlh in #130

New Contributors

• @xrc10 made their first contribution in #14
• @eltociear made their first contribution in #27
• @zhangqianqianhzlh made their first contribution in #49
• @Amos1109 made their first contribution in #65
• @davidluciolu made their first contribution in #58
• @chaoyuhao made their first contribution in #105

Full Changelog: https://github.com/om-ai-lab/VLM-R1/commits/v0.1.0