GH Repository · rllm-org
rllm
Democratizing Reinforcement Learning for LLMs
- stars
- 5,824
- 30-day movement
- starts with the next reading
- Related entries
- 60
- Connections
- 1
pythonPythonverlagent-frameworkdockerllm-trainingdistributed-trainingmachine-learningml-infrastructureagentic-workflowcoding-agentswe-agenttinkerml-platformreinforcement-learningllm-reasoningsearch-agent
rllm is a Python framework for training LLMs with reinforcement learning, described as 'Democratizing Reinforcement Learning for LLMs'. Its topics indicate support for agentic workflows, distributed training, and agent types such as coding, SWE, search, and reasoning agents, with verl listed among related tooling.
It gives you an open-source (Apache-2.0), Python-based starting point for RL training of LLMs on agentic tasks without building the infrastructure yourself.
Use it to
- Train LLMs with reinforcement learning
- Build RL training pipelines for coding or SWE agents
- Run distributed RL training for LLM reasoning
- Prototype search-agent training workflows
For ML engineers and researchers training LLMs with RL
- Role
- agent-framework
- Language
- Python
- Licence
- Apache-2.0
- Forks
- 616
- Open issues
- 78
- Last push
- 2026-09-12
- Latest release
- v0.2.0 · 2025-10-16
topicsreinforcement-learningllm-trainingagent-frameworkdistributed-trainingagentic-workflowpython