Reproduce “Self-ensemble: Mitigating Confidence Mis-calibration for Large Language Models” 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
Zicheng Xu, Guanchu Wang, Guangyao Zheng, Yu-Neng Chuang, Alexander Szalay, Xia Hu, Vladimir Braverman
Self-Ensemble improves multiple-choice question answering by having the same language model evaluate options in randomized smaller groups, average each option’s probabilities, and process groups together with attention masking and positional re-encoding. On QASC, LLaMA-3-8B accuracy reached 80.67%, versus 73.22% standard inference and 75.16% vector scaling. Evidence is limited to tested MCQA settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.