Learning Stateful Predictive Knowledge from Experience

Yan Song, Xidong Feng, Bo Liu, Xinyu Cui, Zichen Liu, Haotian Fu, Mengyue Yang, Cheng Deng, Jian Zhao, Jun Wang

Setup

conda create -n SKL python==3.12
conda activate SKL

pip install -r requirements.txt
cd verl
pip install -e . --no-deps

Running experiments

export MODEL_PATH=xxx
export PYTHONPATH=./

# Create dummy datasets
python tasks/classic_games/chess/data_preprocess/dummy.py

# Run GRPO training
bash experiments/chesspuzzles/run_train_GRPO.sh

# Run NLTS-BFS training
bash experiments/chesspuzzles/run_train_NLTS_BFS_GRPO.sh

Citation

@inproceedings{songlearning,
  title={Learning Stateful Predictive Knowledge From Experience},
  author={Song, Yan and Feng, Xidong and Liu, Bo and Cui, Xinyu and Liu, Zichen and Fu, Haotian and Yang, Mengyue and Deng, Cheng and Zhao, Jian and Wang, Jun},
  booktitle={Second Workshop on Agents in the Wild: Safety, Security, and Beyond}
}