A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Xiaocan, Wu, Shiliang, Shen, Zheng
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908868689264640
author Li, Xiaocan
Wu, Shiliang
Shen, Zheng
author_facet Li, Xiaocan
Wu, Shiliang
Shen, Zheng
contents Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the high data staleness under the asynchronous RL setting. Decoupled loss used in decoupled PPO improves coupled-loss style of algorithms' (e.g., standard PPO, GRPO) learning stability by introducing a proximal policy to decouple the off-policy correction (importance weight) from the policy update constraint (trust region). However, the proximal policy requires an extra forward pass through the model at each training step, creating a computational overhead for large language models training. We observe that since the proximal policy only serves as a trust region anchor between the behavior and target policies, we can approximate it through simple interpolation without explicit computation. We call this approach A-3PO (APproximated Proximal Policy Optimization). A-3PO eliminates this overhead, accelerating training by 1.8x speedup while maintaining comparable performance. Code \& off-the-shelf example are contributed to the open-source RL training system AReaL at: https://github.com/inclusionAI/AReaL/blob/v1.0.0.rc1/docs/algorithms/prox_approx.md
format Preprint
id arxiv_https___arxiv_org_abs_2512_06547
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation
Li, Xiaocan
Wu, Shiliang
Shen, Zheng
Machine Learning
Artificial Intelligence
Distributed, Parallel, and Cluster Computing
Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the high data staleness under the asynchronous RL setting. Decoupled loss used in decoupled PPO improves coupled-loss style of algorithms' (e.g., standard PPO, GRPO) learning stability by introducing a proximal policy to decouple the off-policy correction (importance weight) from the policy update constraint (trust region). However, the proximal policy requires an extra forward pass through the model at each training step, creating a computational overhead for large language models training. We observe that since the proximal policy only serves as a trust region anchor between the behavior and target policies, we can approximate it through simple interpolation without explicit computation. We call this approach A-3PO (APproximated Proximal Policy Optimization). A-3PO eliminates this overhead, accelerating training by 1.8x speedup while maintaining comparable performance. Code \& off-the-shelf example are contributed to the open-source RL training system AReaL at: https://github.com/inclusionAI/AReaL/blob/v1.0.0.rc1/docs/algorithms/prox_approx.md
title A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation
topic Machine Learning
Artificial Intelligence
Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2512.06547