Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor 文章

ArXiv CS.AI2026-06-03NEWSen作者: Xiaocan Li, Shiliang Wu, Zheng Shen
查看原文 →
关系图谱

Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor · 相关人物

暂无数据