OpenThought3-Qwen3-4B

OpenThought3-Qwen3-4B is a math reasoning supervised fine-tuning dataset in chat-message JSONL format.

Data Creation and Cleaning

This dataset was generated by Qwen3-4B (Non-thinking) from math-domain prompts selected from OpenThoughts3-1.2M. The generated responses were cleaned through deduplication, removal of degenerate repetition/repeater-style outputs, and template checks on the assistant responses.

The final cleaned dataset was used to train Qwen3-1.7B-Base, producing Qwen3-1.7B-SFT.

This dataset is linked to:

• Model: https://huggingface.co/lllyx/Qwen3-1.7B-SFT
• Paper: https://huggingface.co/papers/2604.13016
• arXiv: https://arxiv.org/abs/2604.13016

Dataset Structure

Each row is a JSON object with a messages field. The messages value is a list of chat turns, typically containing a user prompt and an assistant response.

Example schema:

{
  "messages": [
    {"role": "user", "content": "..."},
    {"role": "assistant", "content": "..."}
  ]
}

Related Artifact

The dataset is intended for supervised fine-tuning and is associated with lllyx/Qwen3-1.7B-SFT.

Citation

If you use this dataset, please cite:

@article{li2026rethinking,
  title={Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe},
  author={Li, Yaxuan and Zuo, Yuxin and He, Bingxiang and Zhang, Jinqian and Xiao, Chaojun and Qian, Cheng and Yu, Tianyu and Gao, Huan-ang and Yang, Wenkai and Liu, Zhiyuan and Ding, Ning},
  journal={arXiv preprint arXiv:2604.13016},
  year={2026}
}