Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2509.25762 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911487493144576 |
|---|---|
| author | Yan, Kaizhuo Yu, Yingjie Yu, Yifan Zheng, Haizhong Lai, Fan |
| author_facet | Yan, Kaizhuo Yu, Yingjie Yu, Yifan Zheng, Haizhong Lai, Fan |
| contents | Proximal Policy Optimization (PPO)-based reinforcement learning from human feedback (RLHF) is a widely adopted paradigm for aligning large language models (LLMs) with human preferences. However, its training pipeline suffers from substantial inefficiencies due to sequential multi-model dependencies (e.g., reward model depends on actor outputs) and long-tail response lengths, where a few long responses straggle the stage completion. We present OPPO, a novel, lightweight, and model-agnostic PPO-based RLHF framework that improves training efficiency by overlapping pipeline execution. OPPO introduces two novel techniques: (1) Intra-step overlap, which streams upstream model outputs (e.g., actor model) in right-sized chunks, enabling the downstream model (e.g., reward) to begin prefill while the upstream continues decoding; and (2) Inter-step overlap, which adaptively overcommits a few prompts and defers long generations to future steps, mitigating tail latency without discarding partial work. OPPO integrates easily with existing PPO implementations with a lightweight wrapper. Extensive evaluations show that OPPO accelerates PPO-based RLHF training by $1.8\times$--$2.8\times$ and improves GPU utilization by $1.4\times$--$2.1\times$ without compromising training convergence. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_25762 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | OPPO: Accelerating PPO-based RLHF via Pipeline Overlap Yan, Kaizhuo Yu, Yingjie Yu, Yifan Zheng, Haizhong Lai, Fan Machine Learning Proximal Policy Optimization (PPO)-based reinforcement learning from human feedback (RLHF) is a widely adopted paradigm for aligning large language models (LLMs) with human preferences. However, its training pipeline suffers from substantial inefficiencies due to sequential multi-model dependencies (e.g., reward model depends on actor outputs) and long-tail response lengths, where a few long responses straggle the stage completion. We present OPPO, a novel, lightweight, and model-agnostic PPO-based RLHF framework that improves training efficiency by overlapping pipeline execution. OPPO introduces two novel techniques: (1) Intra-step overlap, which streams upstream model outputs (e.g., actor model) in right-sized chunks, enabling the downstream model (e.g., reward) to begin prefill while the upstream continues decoding; and (2) Inter-step overlap, which adaptively overcommits a few prompts and defers long generations to future steps, mitigating tail latency without discarding partial work. OPPO integrates easily with existing PPO implementations with a lightweight wrapper. Extensive evaluations show that OPPO accelerates PPO-based RLHF training by $1.8\times$--$2.8\times$ and improves GPU utilization by $1.4\times$--$2.1\times$ without compromising training convergence. |
| title | OPPO: Accelerating PPO-based RLHF via Pipeline Overlap |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2509.25762 |