Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning

Source

arXiv

arxiv.org/abs/2607.07508v1

Read the source

Source

Plain-Language Brief

In this paper, the authors introduce Single-rollout Asynchronous Optimization (SAO), a new approach to reinforcement learning (RL) for post-training large language models (LLMs). SAO aims to improve training stability and task effectiveness while addressing off-policy challenges in asynchronous RL. It does this by replacing group-wise sampling with single-rollout sampling, using one rollout per prompt, and implementing a strict double-side token-level clipping strategy for optimization stability. The authors claim that SAO outperforms existing methods on various agentic coding and reasoning benchmarks, including SWE-Bench Verified, BeyondAIME, and IMOAnswerBench, and has been successfully deployed in the RL pipeline for training the open GLM-5.2 model.

What the Document Says

  • The paper presents Single-rollout Asynchronous Optimization (SAO) for Reinforcement Learning (RL).
  • SAO is designed to address stability and off-policy challenges in asynchronous RL.
  • Single-rollout sampling replaces group-wise sampling in SAO, using one rollout per prompt.
  • A strict double-side token-level clipping strategy is introduced for optimization stability.
  • SAO outperforms GRPO and its variants on agentic coding and reasoning benchmarks.
  • Examples of these benchmarks include SWE-Bench Verified, BeyondAIME, and IMOAnswerBench.
  • SAO has been successfully deployed in the RL pipeline for training the open GLM-5.2 model.

Key Concepts Mentioned

  • Reinforcement Learning
  • Large Language Models
  • Asynchronous Reinforcement Learning
  • Single-rollout Sampling
  • Double-side Token-level Clipping
  • Stability
  • Off-policy Challenges
  • Agentic Coding and Reasoning Benchmarks
  • GLM-5.2 Model

Claims Made by the Document

  • SAO outperforms GRPO and its variants on agentic coding and reasoning benchmarks.

Evidence Used

  • SAO consistently outperforms GRPO and its variants on agentic coding and reasoning benchmarks such as SWE-Bench Verified, BeyondAIME, and IMOAnswerBench.
  • SAO has been successfully deployed in the RL pipeline for training the open GLM-5.2 model.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Reinforcement learning (RL) is becoming increasingly important for post-training large language models (LLMs). Previous RL pipelines for LLMs were mostly synchronous and batch-interleaved, which is inefficient for long-horizon agentic tasks. Recently, asynchronous RL has emerged as a more efficient alternative by updating the model as rollouts arrive. However, existing asynchronous RL systems often emphasize throughput, while leaving training stability and task effectiveness largely underexplored. For example, a key challenge is that group-wise sampling in the widely-used GRPO framework does not naturally fit asynchronous agentic training. In this paper, we present Single-rollout Asynchronous Optimization (SAO) to address the stability and off-policy challenges in asynchronous RL. To reduce off-policy effects and improve generalization, we replace group-wise sampling with single-rollout sampling, that is, using one rollout per prompt. We further improve this single-rollout strategy with practical value-model training designs. To improve optimization stability, we introduce a strict double-side token-level clipping strategy. SAO is able to train stably for one thousand steps and consistently outperform GRPO and its variants on agentic coding and reasoning benchmarks, such as SWE-Bench Verified, BeyondAIME, and IMOAnswerBench. We also demonstrate that single-rollout RL is particularly effective in a simulated online learning setting, where the model must adapt to changing evolving environments. To this end, SAO is successfully deployed in the agentic RL pipeline for training the open GLM-5.2 model (750B-A40B).