分解 MXFP4 量化误差：为大模型强化学习对症下药
MXFP4 量化误差的精确三分解（缩放偏置 / 死区截断 / 网格噪声）以及对症下药的三件套修正（MBS / OF / AQN）。
MXFP4
Decomposing MXFP4 quantization error for LLM RL: reducible bias, recoverable deadzone, and an irreducible floor
An exact three-way decomposition of MXFP4 quantization error into scale bias, deadzone truncation, and grid noise — and the mechanism-targeted corrections (MBS, OF, AQN) that follow.