Model Card for Qwen3-8B-ReST-RL

This model is trained with the ReST-RL paradigm, based on the Qwen3-8B model. It is trained for 2 reinforce iterations.

Quick start

from transformers import pipeline

question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="None", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Framework versions

• TRL: 0.19.1
• Transformers: 4.51.3
• Pytorch: 2.6.0
• Datasets: 3.2.0
• Tokenizers: 0.21.1

Citations

Cite ReST-RL as:

@misc{zhoubian2025restrlachievingaccuratecode,
      title={ReST-RL: Achieving Accurate Code Reasoning of LLMs with Optimized Self-Training and Decoding}, 
      author={Sining Zhoubian and Dan Zhang and Jie Tang},
      year={2025},
      eprint={2508.19576},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2508.19576}, 
}