OpenManus RL
Basic Information
OpenManus-RL is an open-source project for researching and developing reinforcement learning methods to tune large language model agents. It is a collaborative effort led by Ulab-UIUC and MetaGPT and extends the original OpenManus initiative. The repository aims to explore RL-based paradigms that improve agent reasoning, decision-making, and tool integration by combining trajectory data, reasoning models, and RL tunning techniques. It aggregates agent trajectory datasets, outlines methodologies for reward modeling and rollout strategies, and provides a simplified library for supervised fine-tuning and generalized reward-based policy optimization built atop existing RL toolkits. The project is maintained as a live-streaming research effort with planned code and dataset releases, benchmark evaluations on agent benchmarks, and an invitation for community contributions.
Links
App Details
-
$$$$ -
$$$$