Reproduce “Neurons Are All You Need” 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
Joseph B. Alzagatiti, Alain C. Glanzman, Perry A. Demsko, Pablo I. Salanova, David L. Glanzman
NURESA transfers selected memory from a fine-tuned model to a fresh matching model without further recipient training. Its unit-level registers accumulate activation history, and committed states are packaged separately from ordinary weights. In zero-shot Qwen2.5-1.5B GSM8K evaluation, transfer reached 55.8% accuracy, versus 37.1% for base and 63.3% for math fine-tuning. The result demonstrates transfer only in tested same-architecture configurations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.