# TIGER-AI-Lab/verl-tool

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

Repository: https://github.com/TIGER-AI-Lab/verl-tool
Canonical: https://ross.abutalabs.com/products/verl-tool
Homepage: https://arxiv.org/pdf/2509.01055
Language: Python
License: MIT
License Family: permissive
Topics: learning, llm, reinforcement, agent
Last push: 2026-07-15T18:11:03+00:00

## Health v2 (maintenance only)
Score: 66/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 92, release rhythm 50, longevity 37
- inputs: {"age_days": 530, "days_push": 49, "days_rel": 252, "gap_med": 43, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1036, forks 89 (observed 2026-08-28T04:03:19.376877+00:00)

## What it is
VerlTool is a unified, extensible framework built on verl for training LLM agents with tool use via reinforcement learning. It decouples actor rollout generation from environment interaction and ships training recipes like Search-R1, ToRL, and DAPO.

## Use cases
- train an LLM agent to call tools with reinforcement learning
- reproduce Search-R1 style search-augmented RL training
- run asynchronous trajectory-level RL rollouts with tool calling
- train models for NL2SQL with tool feedback
- add custom tool environments to an RL training pipeline
- benchmark tool-using LLM agents

## When to choose
- you want to RL-train LLMs that interact with external tools or environments
- you already use verl and need tool-call rollout support
- you need reproducible recipes like Search-R1, ToRL, or DAPO

## When to avoid
- you only need inference-time agent orchestration without RL training
- you need a simple SFT-only fine-tuning pipeline
- you lack GPU resources for large-scale RL training

## Facets
- artifact type: framework
- maturity: active
- function: reinforcement-learning, llm-training, agent-framework, machine-learning
- domain: reinforcement-learning, large-language-models, machine-learning
- platform: python
- tags: tool-use, rl-training, verl, llm-agents, rollout, tool-agent, research, ai-agents, gpu, linux, docker

## Member repositories
- TIGER-AI-Lab/verl-tool (main) score 66

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:19.376877+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:04:28.679495+00:00, confidence not recorded.
  - readme: https://github.com/TIGER-AI-Lab/verl-tool (fetched 2026-08-28T04:03:19.376877+00:00, sha 9a29951373a0)
- Data as of 2026-08-30T08:39:29.467469+00:00.