English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

SPO++: Stream-Aligned Policy Optimization for Asynchronous Agentic RL

Forum topic · 小凯 · 2026-08-27

Summary

SPO++ is a new reinforcement learning method for asynchronous agentic RL that builds on Single-stream Policy Optimization (SPO). Group-relative RL requires waiting for sibling rollouts of the same prompt, which is costly for long, variable tool-use trajectories. SPO removes this dependency using a persistent prompt-level value estimate, but it whitens one advantage per trajectory before optimizing a token-mean actor loss. The authors show that trajectory centering generally does not center the token-weighted quantity consumed by the actor, and fix this mismatch by standardizing terminal-outcome advantages under the action-token measure. SPO++ additionally organizes prompt evidence by the policy event that generated it rather than by learner receipt order. In matched runs on ALFWorld at two model scales and on Math-TIR, SPO++ improves online learning efficiency over SPO, with pairwise ablations identifying action-token-measure normalization as the strongest component. Paper: arXiv 2608.24870.

Paper Overview

Field: AI Authors: Kai Ruan, Jinghao Lin, Qianshan Wei, Ziqi Zhou, Zihe Huang Published: 2026-08-25 arXiv: 2608.24870

Summary

Group-relative reinforcement learning waits for sibling rollouts of the same prompt, which is costly for long and variable tool-use trajectories. Single-stream Policy Optimization (SPO) removes this dependency with a persistent prompt-level value estimate, but its recipe whitens one advantage per trajectory before optimizing a token-mean actor loss.

The authors show that trajectory centering generally does not center the token-weighted quantity consumed by the actor, and fix this mismatch by standardizing terminal-outcome advantages under the action-token measure. In addition, SPO++ organizes prompt evidence by the policy event that generated it rather than by learner receipt order.

Results

  • Evaluated on ALFWorld at two model scales and on Math-TIR in matched runs.
  • SPO++ improves online learning efficiency over SPO.
  • Pairwise ablation experiments identify action-token-measure normalization as the strongest tested component.

Original Abstract

Group-relative reinforcement learning waits for sibling rollouts of the same prompt, which is costly for long and variable tool-use trajectories. Single-stream Policy Optimization (SPO) removes this dependency with a persistent prompt-level value estimate, but its recipe whitens one advantage per trajectory before optimizing a token-mean actor loss. We show that trajectory centering generally does not center the token-weighted quantity consumed by the actor, and fix the mismatch by standardizing terminal-outcome advantages under the action-token measure. We additionally organize prompt evidence by the policy event that generated it rather than learner receipt order. Across matched runs on ALFWorld at two model scales and on Math-TIR, SPO++ improves online learning efficiency over SPO.

---

*Auto-collected on 2026-08-27*

Tags

#reinforcement-learning#agentic-rl#policy-optimization#llm-training#arxiv#ai-research

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178634086