quick start

ENVIRONMENT

conda create -n Alpha python=3.10
conda activate Alpha
pip install -r requirements.txt
wandb login

scripts prepare and training

Open ./examples/ppo_trainer/qwen3B-instruct.sh and fill your checkpoint save dir, and start running qwen3B-instruct.

bash ./examples/ppo_trainer/qwen3B-instruct.sh

model merge

Open merge script, filled in your specific checkpoint path and hf_model_path to get the final model

More Details

please refer to Verl Official readme

Evaluation

Open ./Evaluation