Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wen, Xumeng, Liu, Zihan, Zheng, Shun, Ye, Shengyu, Wu, Zhirong, Wang, Yang, Xu, Zhijian, Liang, Xiao, Li, Junjie, Miao, Ziming, Bian, Jiang, Yang, Mao
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911188380549120
author Wen, Xumeng
Liu, Zihan
Zheng, Shun
Ye, Shengyu
Wu, Zhirong
Wang, Yang
Xu, Zhijian
Liang, Xiao
Li, Junjie
Miao, Ziming
Bian, Jiang
Yang, Mao
author_facet Wen, Xumeng
Liu, Zihan
Zheng, Shun
Ye, Shengyu
Wu, Zhirong
Wang, Yang
Xu, Zhijian
Liang, Xiao
Li, Junjie
Miao, Ziming
Bian, Jiang
Yang, Mao
contents Recent advancements in long chain-of-thought (CoT) reasoning, particularly through the Group Relative Policy Optimization algorithm used by DeepSeek-R1, have led to significant interest in the potential of Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Models (LLMs). While RLVR promises to improve reasoning by allowing models to learn from free exploration, there remains debate over whether it truly enhances reasoning abilities or simply boosts sampling efficiency. This paper systematically investigates the impact of RLVR on LLM reasoning. We revisit Pass@K experiments and demonstrate that RLVR can extend the reasoning boundary for both mathematical and coding tasks. This is supported by our introduction of a novel evaluation metric, CoT-Pass@K, which captures reasoning success by accounting for both the final answer and intermediate reasoning steps. Furthermore, we present a theoretical framework explaining RLVR's incentive mechanism, demonstrating how it can encourage correct reasoning even when rewards are based solely on answer correctness. Our analysis of RLVR's training dynamics reveals that it incentivizes correct reasoning early in the process, with substantial improvements in reasoning quality confirmed through extensive evaluations. These findings provide strong evidence of RLVR's potential to enhance LLM reasoning, offering valuable insights into its mechanisms and performance improvements.
format Preprint
id arxiv_https___arxiv_org_abs_2506_14245
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs
Wen, Xumeng
Liu, Zihan
Zheng, Shun
Ye, Shengyu
Wu, Zhirong
Wang, Yang
Xu, Zhijian
Liang, Xiao
Li, Junjie
Miao, Ziming
Bian, Jiang
Yang, Mao
Artificial Intelligence
Computation and Language
Recent advancements in long chain-of-thought (CoT) reasoning, particularly through the Group Relative Policy Optimization algorithm used by DeepSeek-R1, have led to significant interest in the potential of Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Models (LLMs). While RLVR promises to improve reasoning by allowing models to learn from free exploration, there remains debate over whether it truly enhances reasoning abilities or simply boosts sampling efficiency. This paper systematically investigates the impact of RLVR on LLM reasoning. We revisit Pass@K experiments and demonstrate that RLVR can extend the reasoning boundary for both mathematical and coding tasks. This is supported by our introduction of a novel evaluation metric, CoT-Pass@K, which captures reasoning success by accounting for both the final answer and intermediate reasoning steps. Furthermore, we present a theoretical framework explaining RLVR's incentive mechanism, demonstrating how it can encourage correct reasoning even when rewards are based solely on answer correctness. Our analysis of RLVR's training dynamics reveals that it incentivizes correct reasoning early in the process, with substantial improvements in reasoning quality confirmed through extensive evaluations. These findings provide strong evidence of RLVR's potential to enhance LLM reasoning, offering valuable insights into its mechanisms and performance improvements.
title Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2506.14245