Tag
This paper from Tsinghua University introduces Single-rollout Asynchronous Optimization (SAO) for reinforcement learning post-training of LLMs. SAO replaces batch-based GRPO with single-rollout asynchronous training to reduce idle GPU time and improve stability, and it was used to train the GLM-5.2 model (750B-A40B), achieving state-of-the-art results on agentic coding and reasoning benchmarks.
The paper introduces Single-rollout Asynchronous Optimization (SAO) to address stability and off-policy challenges in asynchronous RL for LLM post-training, and demonstrates that SAO consistently outperforms GRPO on agentic coding and reasoning benchmarks.
Tsinghua's SAO algorithm addresses stability and off-policy drift in asynchronous reinforcement learning for LLMs, achieving consistent improvements over GRPO on agentic coding and reasoning benchmarks, and is used to train the GLM-5.2 model.
Prime Intellect releases prime-rl v0.6.0, enabling efficient reinforcement learning at trillion-parameter scale on large Mixture-of-Experts models, with sub-5-minute step times and optimizations for asynchronous RL.
A blog post exploring weight synchronization techniques in asynchronous reinforcement learning, covering transport and payload trade-offs across frameworks.
Luke J. Huang's new blog post surveys asynchronous reinforcement learning theory and infrastructure across 8 open-weight frontier labs, addressing algorithmic techniques and systems fixes for train-inference mismatch.