zeweizhang/ImitSAT-KeyTrace

ImitSAT‑KeyTrace (Download & Preview)

     

Summary. This repository hosts the bucketed training data for ImitSAT, a learned branching policy for CDCL SAT solving, presented in the paper Boolean Satisfiability via Imitation Learning.
Each bucket is a gzip‑compressed JSON Lines file (*.jsonl.gz) of KeyTrace training examples for random 3‑SAT with planted assignments.

⚠️ Use our code to load & preprocess. We intentionally do not provide a datasets.load_dataset builder here.
The official loader is in the GitHub repo (see ImitSAT/ImitSAT_dataset.py). Point your training config’s dataset_path to the downloaded folder.

Download

git lfs install
git clone https://huggingface.co/datasets/zeweizhang/ImitSAT-KeyTrace

• sat_5_15.jsonl.gz
• sat_16_30.jsonl.gz
• sat_31_60.jsonl.gz
• sat_61_100.jsonl.gz
• sat_50_50.jsonl.gz
• sat_100_100.jsonl.gz

Sample Usage

This dataset is designed to be used with the ImitSAT code repository for training. The official loader for this dataset is ImitSAT/ImitSAT_dataset.py within the ImitSAT GitHub repository.

Environment Setup

First, set up the required Python environment as specified in the GitHub repository:

conda create -n imitsat python=3.10 -y
conda activate imitsat
pip install -r requirements.txt

Training ImitSAT

After setting up the environment and downloading the ImitSAT-KeyTrace dataset (as shown in the Download section above), you can train the ImitSAT model by pointing your training configuration's dataset_path to the downloaded folder. An example training command from the ImitSAT repository is:

python  train_ImitSAT.py  --config ./model_config/ImitSAT_config.json
# (Optional) resume from a folder containing ImitSAT.npz and tokenizer/
python  train_ImitSAT.py  --config ./model_config/ImitSAT_config.json  --resume_dir ./model_ckpt/

Preview

The Dataset Viewer shows only small, human‑readable files under preview/ (100 examples each)

Citation

@misc{zhang2025booleansatisfiabilityimitationlearning,
      title={Boolean Satisfiability via Imitation Learning}, 
      author={Zewei Zhang and Huan Liu and Yuanhao Yu and Jun Chen and Xiangyu Xu},
      year={2025},
      eprint={2509.25411},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2509.25411}, 
}