verl-recipe

verl-recipe hosts recipes based on verl contributed by the community.

Usage

verl-recipe can be used as a submodule of verl, keeping backward compatibility as verl/recipe:

git clone https://github.com/verl-project/verl.git
cd verl
git submodule update --init --recursive recipe

Available Recipes

• retool: Reinforcement Learning for Strategic Tool Use in LLMs
• langgraph_agent: A tiny example to demonstrate multi-turn rollout with LangGraph ReactAgent to solve math expression.
• spo: Single-stream Policy Optimization.
• TBA...

Contribution

Version Specification

Recipes are recommended to specify the verl version required, e.g.,

# release version
verl==0.6.0

# dev version
verl@git+https://github.com/volcengine/verl.git@313dfdb2199124a37189e32e6d4a6c654379f2d4

Code Linting and Formatting

To maximize flexiblility but minimize meaningless changes, we apply pre-commit but only force code linting and formatting with ruff. Use it as follows:

pip install pre-commit
pre-commit install
# for staged changes
pre-commit run
# for all files in the repo
pre-commit run --all-files