rLLM
Reinforcement-learning framework that reuses agent harnesses for evaluation and training, collecting rollouts and rewards across sandbox backends.
Tags
Description
Python 3.11+ source install via uv pip install "rllm @ git+https://github.com/rllm-org/rllm.git". CLI entry points include rllm eval and rllm train; @rllm.rollout wraps custom agents. The framework integrates CLI harnesses, Harbor task directories, Docker/Daytona/Modal/local execution and verl, Tinker and Fireworks training backends. Model gateway traces feed reward computation and policy updates. Apache-2.0 software; GPU, model and third-party platform services are separate.
Agent access
Related
- verl: Distributed reinforcement-learning training library for LLMs, with GRPO/PPO dataflows and pluggable training and rollout engines for agentic post-training.
- Harbor: Framework from the Terminal-Bench creators for evaluating and optimizing agents and models in sandboxed environments, run in parallel across cloud sandbox providers.
- Terminal-Bench: Continuously maintained benchmark and leaderboard (Stanford, Laude Institute) measuring how well AI agents complete hard, end-to-end tasks in real terminal environments.
Sources
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)