Process Reinforcement through Implicit Rewards
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908559670771712 |
|---|---|
| author | Cui, Ganqu Yuan, Lifan Wang, Zefan Wang, Hanbin Zhang, Yuchen Chen, Jiacheng Li, Wendi He, Bingxiang Fan, Yuchen Yu, Tianyu Xu, Qixin Chen, Weize Yuan, Jiarui Chen, Huayu Zhang, Kaiyan Lv, Xingtai Wang, Shuo Yao, Yuan Han, Xu Peng, Hao Cheng, Yu Liu, Zhiyuan Sun, Maosong Zhou, Bowen Ding, Ning |
| author_facet | Cui, Ganqu Yuan, Lifan Wang, Zefan Wang, Hanbin Zhang, Yuchen Chen, Jiacheng Li, Wendi He, Bingxiang Fan, Yuchen Yu, Tianyu Xu, Qixin Chen, Weize Yuan, Jiarui Chen, Huayu Zhang, Kaiyan Lv, Xingtai Wang, Shuo Yao, Yuan Han, Xu Peng, Hao Cheng, Yu Liu, Zhiyuan Sun, Maosong Zhou, Bowen Ding, Ning |
| contents | Dense process rewards have proven a more effective alternative to the sparse outcome-level rewards in the inference-time scaling of large language models (LLMs), particularly in tasks requiring complex multi-step reasoning. While dense rewards also offer an appealing choice for the reinforcement learning (RL) of LLMs since their fine-grained rewards have the potential to address some inherent issues of outcome rewards, such as training efficiency and credit assignment, this potential remains largely unrealized. This can be primarily attributed to the challenges of training process reward models (PRMs) online, where collecting high-quality process labels is prohibitively expensive, making them particularly vulnerable to reward hacking. To address these challenges, we propose PRIME (Process Reinforcement through IMplicit rEwards), which enables online PRM updates using only policy rollouts and outcome labels through implict process rewards. PRIME combines well with various advantage functions and forgoes the dedicated reward model training phrase that existing approaches require, substantially reducing the development overhead. We demonstrate PRIME's effectiveness on competitional math and coding. Starting from Qwen2.5-Math-7B-Base, PRIME achieves a 15.1% average improvement across several key reasoning benchmarks over the SFT model. Notably, our resulting model, Eurus-2-7B-PRIME, surpasses Qwen2.5-Math-7B-Instruct on seven reasoning benchmarks with 10% of its training data. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_01456 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Process Reinforcement through Implicit Rewards Cui, Ganqu Yuan, Lifan Wang, Zefan Wang, Hanbin Zhang, Yuchen Chen, Jiacheng Li, Wendi He, Bingxiang Fan, Yuchen Yu, Tianyu Xu, Qixin Chen, Weize Yuan, Jiarui Chen, Huayu Zhang, Kaiyan Lv, Xingtai Wang, Shuo Yao, Yuan Han, Xu Peng, Hao Cheng, Yu Liu, Zhiyuan Sun, Maosong Zhou, Bowen Ding, Ning Machine Learning Artificial Intelligence Computation and Language Dense process rewards have proven a more effective alternative to the sparse outcome-level rewards in the inference-time scaling of large language models (LLMs), particularly in tasks requiring complex multi-step reasoning. While dense rewards also offer an appealing choice for the reinforcement learning (RL) of LLMs since their fine-grained rewards have the potential to address some inherent issues of outcome rewards, such as training efficiency and credit assignment, this potential remains largely unrealized. This can be primarily attributed to the challenges of training process reward models (PRMs) online, where collecting high-quality process labels is prohibitively expensive, making them particularly vulnerable to reward hacking. To address these challenges, we propose PRIME (Process Reinforcement through IMplicit rEwards), which enables online PRM updates using only policy rollouts and outcome labels through implict process rewards. PRIME combines well with various advantage functions and forgoes the dedicated reward model training phrase that existing approaches require, substantially reducing the development overhead. We demonstrate PRIME's effectiveness on competitional math and coding. Starting from Qwen2.5-Math-7B-Base, PRIME achieves a 15.1% average improvement across several key reasoning benchmarks over the SFT model. Notably, our resulting model, Eurus-2-7B-PRIME, surpasses Qwen2.5-Math-7B-Instruct on seven reasoning benchmarks with 10% of its training data. |
| title | Process Reinforcement through Implicit Rewards |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2502.01456 |