BigHugger
GH Repository · TIGER-AI-Lab

verl-tool

A version of verl to support diverse tool use [TMLR 2026]

stars
1,041
30-day movement
+31/day
Related entries
60
Connections
1
agent-appPythonagentreinforcementllmpythonlearning

verl-tool is a modified version of the verl reinforcement learning framework that adds support for diverse tool use during agent training. It is published as Python code under an MIT license by TIGER AI Lab, with an associated TMLR 2026 paper.

Use it when you want to train LLM agents with reinforcement learning in verl while allowing them to call external tools.

Use it to

  • Train LLM agents with tool-calling via RL
  • Extend verl pipelines with custom tools
  • Run reinforcement learning experiments on agentic tasks
  • Reproduce research from the accompanying TMLR paper

For Researchers and engineers training tool-using LLM agents

Role
agent-app
Language
Python
Licence
MIT
Forks
90
Open issues
8
Last push
2026-07-15
Latest release
v0.1.0 · 2025-11-10
topicsreinforcement-learningllmagentstool-usetrainingresearch