README.md

metadata

task_categories:
  - other

ReproduceBench

ReproduceBench is a benchmark featuring verified implementations and comprehensive metrics for evaluating both reproduction and execution fidelity of AI experiments. It was introduced in the paper AutoReproduce: Automatic AI Experiment Reproduction with Paper Lineage.

GitHub Repository: https://github.com/AI9Stars/AutoReproduce

Download Datasets

All the datasets and human-curated reference code are available here. You can download the dataset using the Hugging Face CLI:

pip install -U huggingface_hub
huggingface-cli download --repo-type dataset --resume-download ai9stars/ReproduceBench --local-dir ReproduceBench

Data Preparation

All the datasets are included in the reproducebench.tar.gz. You need firstly download and decompress the data. The overall file structure after decompressing is as follows:

PreproduceBench
|-- PreproduceBench
    |-- itransformer
        |-- source/ # contain the dataset
        |-- dataloader.py
        |-- run_itransformer.py
    |-- lsm
        |-- source/ # contain the dataset
        |-- dataloader.py
        |-- run_itransformer.py
    ...

Citation

If you find this work useful, please cite our paper:

@misc{zhao2025autoreproduceautomaticaiexperiment,
      title={AutoReproduce: Automatic AI Experiment Reproduction with Paper Lineage}, 
      author={Xuanle Zhao and Zilin Sang and Yuxuan Li and Qi Shi and Shuo Wang and Duzhen Zhang and Xu Han and Zhiyuan Liu and Maosong Sun},
      year={2025},
      eprint={2505.20662},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2505.20662}, 
}