Chen-Liang Li

Where Does Staleness Accumulate? Pool Aware Effective Staleness Control for Asynchronous RL in LLM Post-Training

This work decomposes trajectory staleness into Generation Staleness, accumulated before rollout completion, and Waiting Staleness, accumulated after a completed trajectory enters the pool, and introduces PACE (Pool-Aware Control of Effective Staleness), which improves validation performance in multi-turn tool-integrate...