---
license: mit
pipeline_tag: text-generation
library_name: transformers
---

# Co-rewarding: Stable Self-supervised RL for Eliciting Reasoning in Large Language Models

This is the **TMLR-Group-HF/Entropy-Qwen3-4B-Base** model. It is the Qwen3-4B-Base model trained by the Entropy Minimization method using the MATH training set, as presented in the paper [Co-rewarding: Stable Self-supervised RL for Eliciting Reasoning in Large Language Models](https://huggingface.co/papers/2508.00410).

For more details on the Co-rewarding framework, code, and other checkpoints, please refer to the official GitHub repository: [https://github.com/tmlr-group/Co-rewarding](https://github.com/tmlr-group/Co-rewarding).

## Abstract
While reinforcement learning with verifiable rewards (RLVR) is effective to improve the reasoning ability of large language models (LLMs), its reliance on human-annotated labels leads to the scaling up dilemma, especially for complex tasks. Recent self-rewarding methods investigate a label-free alternative to unlock the reasoning capabilities of LLMs, yet they frequently encounter the non-negligible training collapse issue, as the single-view supervision signal easily forms the self-consistent illusion, yielding the reward hacking. Inspired by the success of self-supervised learning, we propose *Co-rewarding*, a novel self-supervised RL framework that improves training stability by seeking complementary supervision from another views. Specifically, we instantiate Co-rewarding in two ways: (1) *Co-rewarding-I* is a data-side instantiation that derives reward signals from contrastive agreement across semantically analogous questions; and (2) *Co-rewarding-II* is a model-side instantiation that maintains a slowly-updated reference teacher with pseudo labels to realize self-distillation. Intuitively, such instantiations introduce different levels of discrepancy to increase the difficulty of training collapse on trivial reasoning solutions. Empirically, Co-rewarding exhibits stable training across various setups, and outperforms other self-rewarding baselines by +3.31% improvements on average on multiple mathematical reasoning benchmarks, especially by +7.49% on Llama-3.2-3B-Instruct. Notably, Co-rewarding reaches or even surpasses RLVR with ground-truth (GT) label in several cases, such as a Pass@1 of 94.01% on GSM8K with Qwen3-8B-Base remarkably higher than GT.

## Citation
If you use our datasets or models, please cite our paper!
```bibtex
@article{zhang2025co,
  title={Co-rewarding: Stable Self-supervised RL for Eliciting Reasoning in Large Language Models},
  author={Zhang, Zizhuo and Zhu, Jianing and Ge, Xinmu and Zhao, Zihua and Zhou, Zhanke and Li, Xuan and Feng, Xiao and Yao, Jiangchao and Han, Bo},
  journal={arXiv preprint arXiv:2508.00410},
  year={2025}
}
```