Mixture-of-Experts (MoE) models scale capacity via sparse activation but stress memory and bandwidth. Offloading alleviates GPU memory by fetching experts on demand, yet token-level routing causes irregular transfers that make inference I/O-bound. Static uniform quantization reduces traffic but degrades accuracy under aggressive compression by ignoring expert heterogeneity. We present Bandwidth-Efficient Adaptive Mixture-of-Experts via Low-Rank Compensation, which performs router-guided precision restoration using precomputed low-rank compensators. At inference time, our method transfers compact low-rank factors with Top-n (n<k) experts per token and applies compensation to them, keeping others low-bit. Integrated with offloading on GPU and GPU-NDP systems, our method delivers a superior bandwidth-accuracy trade-off and improved throughput.
@article{arxiv.2512.17073,
title = {Bandwidth-Efficient Adaptive Mixture-of-Experts via Low-Rank Compensation},
author = {Zhenyu Liu and Yunzhen Liu and Zehao Fan and Garrett Gagnon and Yayue Hou and Nan Wu and Yangwook Kang and Liu Liu},
journal= {arXiv preprint arXiv:2512.17073},
year = {2025}
}