SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913020975775744 |
|---|---|
| author | Wang, Tianyi Li, Yixia Li, Long Chen, Yibiao Huang, Shaohan Chen, Yun Li, Peng Liu, Yang Chen, Guanhua |
| author_facet | Wang, Tianyi Li, Yixia Li, Long Chen, Yibiao Huang, Shaohan Chen, Yun Li, Peng Liu, Yang Chen, Guanhua |
| contents | Proximal Policy Optimization (PPO) is central to aligning Large Language Models (LLMs) in reasoning tasks with verifiable rewards. However, standard token-level PPO struggles in this setting due to the instability of temporal credit assignment over long Chain-of-Thought (CoT) horizons and the prohibitive memory cost of the value model. While critic-free alternatives like GRPO mitigate these issues, they incur significant computational overhead by requiring multiple samples for baseline estimation, severely limiting training throughput. In this paper, we introduce Sequence-Level PPO (SPPO), a scalable algorithm that harmonizes the sample efficiency of PPO with the stability of outcome-based updates. SPPO reformulates the reasoning process as a Sequence-Level Contextual Bandit problem, employing a decoupled scalar value function to derive low-variance advantage signals without multi-sampling. Extensive experiments on mathematical benchmarks demonstrate that SPPO significantly surpasses standard PPO and matches the performance of computation-heavy group-based methods, offering a resource-efficient framework for aligning reasoning LLMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_08865 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks Wang, Tianyi Li, Yixia Li, Long Chen, Yibiao Huang, Shaohan Chen, Yun Li, Peng Liu, Yang Chen, Guanhua Artificial Intelligence Proximal Policy Optimization (PPO) is central to aligning Large Language Models (LLMs) in reasoning tasks with verifiable rewards. However, standard token-level PPO struggles in this setting due to the instability of temporal credit assignment over long Chain-of-Thought (CoT) horizons and the prohibitive memory cost of the value model. While critic-free alternatives like GRPO mitigate these issues, they incur significant computational overhead by requiring multiple samples for baseline estimation, severely limiting training throughput. In this paper, we introduce Sequence-Level PPO (SPPO), a scalable algorithm that harmonizes the sample efficiency of PPO with the stability of outcome-based updates. SPPO reformulates the reasoning process as a Sequence-Level Contextual Bandit problem, employing a decoupled scalar value function to derive low-variance advantage signals without multi-sampling. Extensive experiments on mathematical benchmarks demonstrate that SPPO significantly surpasses standard PPO and matches the performance of computation-heavy group-based methods, offering a resource-efficient framework for aligning reasoning LLMs. |
| title | SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2604.08865 |