TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement Learning
arXiv:2610.07043v2 Announce Type: replace-cross Abstract: Low-precision execution can substantially accelerate reinforcement learning (RL) for large language models, but discrepancies between learner and sampler execution can destabilize policy optimization. In this paper, we characterize the interaction between mismatch and the policy-gradient direction, distinguishing locally amplifying from contracting update contributions that mismatch magnitude alone cannot identify. In native NVFP4 runs,…