Archives

2026  2

May  2

分解 MXFP4 量化误差：为大模型强化学习对症下药

May 27, 2026 · 5 min · 1023 words · Xiaocan (Bruce) Li

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

May 27, 2026 · 11 min · 2203 words · Xiaocan (Bruce) Li

2025  1

February  1

A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation

February 5, 2025 · 8 min · 1585 words · Xiaocan (Bruce) Li