作者位次 #10/12ACL 2026Findings· Findings of ACL
Q a RL : Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training–Inference Mismatch
Hao Gu · Hao Wang · Jiacheng Liu · Lujun Li · Qiyuan Zhu · Bei Liu · Binxing Xu · Lei Wang · Xintong Yang · Sida Lin · Sirui Han · Yike Guo
摘要
Large language model (LLM) reinforcement learning (RL) pipelines are often bottlenecked by rollout generation, making end-to-end training slow. Recent work mitigates this by running rollouts with quantization to accelerate decoding, which is the most expensive stage of the RL loop. However, these...