Reproduce “MetaEncoder: Exploring the Limit of Bi-Encoders for Multimodal System One Decision Making with Natural Language Interface” 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
Jianpeng Cheng, Guangyu Sun, Aashu Singh, Benyu Zhang, Haixing Dai, Hossein Mansour, Jiangfan Zhang, Shlok Kumar Mishra and 6 more
MetaEncoder tests bounded choice-making by matching separately encoded candidates to a request. It is a 30-billion-parameter multimodal bi-encoder and puts small option lists in requests while scoring candidates by cosine similarity. Option lists raised unweighted subject means from 0.5645 to 0.7456 on MMLU and from 0.4875 to 0.5629 on MMMU; these results concern small closed-set choices, not deployment speed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.