Reproduce “Memorilla: Latent Semantic Memory for 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
Adibvafa Fallahpour, Parsa Idehpour, Vignesh Kothapalli, Nikita Mounier, Shirley Wu, Shayan Pardis, Jure Leskovec
Memorilla lets frozen LLMs consult document stores through reusable embeddings. A learned, query-conditioned reader attends across stored embeddings and supplies the frozen model with 16 memory tokens. On FactKG, it reaches 84.33% accuracy versus 80.43% for full-context inference, using 108× fewer document tokens; across eight benchmarks, it leads five but does not lead on PubMedQA.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.