MMODELYST
People/Shiliang Wu
PPL

Shiliang Wu

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