MMODELYST
People/Xiaocan Li
PPL

Xiaocan Li

Papers (1)
Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor