#5252nd of 60 in Training and fine-tuning
easyr1
RL post-training framework for LLMs and vision-language models, forked from veRL
- Stars
- 5.2k
- License
- Apache-2.0
- Last commit
- Nov 2026
- Last release
- Sep 2025
- Language
- Python
4,830 on 21 Apr 20265,214 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
EasyR1 is a fork of veRL that adds vision-language model support to RL post-training. It runs GRPO, DAPO, Reinforce++, ReMax, RLOO, GSPO and CISPO on Llama3, Qwen2 to Qwen3 and Qwen-VL models, using vLLM for rollouts and Ray for multi-node runs. Example scripts cover Geometry3K, with full and LoRA training and checkpoint merging to Hugging Face format.
Who it is for: ML engineers running RL post-training on LLMs and VLMs with GPU clusters
Strengths
- Supports GRPO, DAPO, Reinforce++, ReMax, RLOO, GSPO and CISPO in one codebase
- Handles Qwen2-VL, Qwen2.5-VL and Qwen3-VL alongside text-only LLMs
- README gives estimated GPU counts per model size from 1.5B to 72B
- Ray multi-node setup documented for 70B+ models; LoRA and padding-free training supported
Weaknesses
- Ulysses parallelism does not work with vision-language models yet
- No supervised fine-tuning or inference scripts; README points to LlamaFactory
- Needs multiple GPUs for most model sizes; hardware figures are estimates
- Pins a specific stack: flash-attn, vLLM 0.8.3+, transformers 4.54+
What it needs
- GPU required
- Docker
- Needs Ray, vLLM, flash-attn, transformers, Docker or Apptainer (optional)
- Models: Llama3, Qwen2, Qwen2.5, Qwen3, Qwen2-VL
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Training and fine-tuning
See all 60| Rank | Project | Score |
|---|---|---|
| 1 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 89 out of 100 |
| 2 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 3 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 87 out of 100 |
| #4 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 86 out of 100 |
| #5 | megatron-lmDistributed training library for large transformer models on NVIDIA GPUs | 85 out of 100 |