#artificial intelligence
Nov 2025
Dynamic Expert Quantization for Scalable Mixture-of-Experts Inference
DynaExq is presented, a runtime-aware mixed-precision serving system that treats single-GPU MoE inference under a hard HBM envelope as an online, budget-constrained precision allocation problem and keeps the experts that dominate runtime traffic resident at higher precision.
Kexin Chu, Dawei Xiang, Zixu Shen et al.
· arXiv.org · 10 citations