VerlTool
Holistic RL training framework for tool-using language agents. Extends the verl framework with multi-turn rollout, tool execution, and reward integration for agentic tasks.
VerlTool is an open-source framework for holistic reinforcement learning training of tool-using language agents. Built on top of the verl framework, it supports multi-turn rollout with real tool execution, flexible reward shaping, and scalable training pipelines.
Key features:
- Multi-turn agent rollout with live tool calls (code interpreter, web search, APIs)
- Modular reward functions for tool-use fidelity and task completion
- Compatible with major open LLMs (Qwen, Llama, Mistral, etc.)
- Efficient distributed training via FSDP + vLLM