MusicPile is a large-scale music-language pre-training dataset jointly launched by the Multimodal Art Projection Research Community, Skywork AI, and the Hong Kong University of Science and Technology. The dataset contains 5.17 million samples and approximately 4.16 billion tokens, from sources including online corpora, encyclopedias, music books, YouTube music subtitles, ABC notation works, mathematical content, and code. The dataset contains three fields: id, text, and src, and each text has no more than 2,048 tokens. MusicPile covers a wide range of music common sense, knowledge questions and answers, and typical music theory content, which plays a key role in improving the music understanding and creation capabilities of large models.

Citation

@misc{yuan2024chatmusician,
title={ChatMusician: Understanding and Generating Music Intrinsically with LLM},
author={Ruibin Yuan and Hanfeng Lin and Yi Wang and Zeyue Tian and Shangda Wu and Tianhao Shen and Ge Zhang and Yuhang Wu and Cong Liu and Ziya Zhou and Ziyang Ma and Liumeng Xue and Ziyu Wang and Qin Liu and Tianyu Zheng and Yizhi Li and Yinghao Ma and Yiming Liang and Xiaowei Chi and Ruibo Liu and Zili Wang and Pengfei Li and Jingcheng Wu and Chenghua Lin and Qifeng Liu and Tao Jiang and Wenhao Huang and Wenhu Chen and Emmanouil Benetos and Jie Fu and Gus Xia and Roger Dannenberg and Wei Xue and Shiyin Kang and Yike Guo},
year={2024},
eprint={2402.16153},
archivePrefix={arXiv},
primaryClass={cs.SD}
}