Reproduce “An Empirical Study of Parameter-Efficient Fine-Tuning for Mixture-of-Experts LLMs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Minglei Li, Peng Ye, Haonan He, Jingqi Ye, Qingyu Yang, Tao Chen
The authors compare parameter-efficient adaptation methods for mixture-of-experts language models, focusing on low-rank updates inside existing expert weights. They evaluate three backbones on two downstream tasks and measure retained general capabilities. Rank-one expert-weight LoRA averages 68.5% downstream, near rank-256 module bypass at 68.4%, with smaller measured capability changes; these are study-specific averages, not a universal guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.