Reproduce “Latest Exact Match Attention” 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
Moritz Brösamle
LEMA turns autoregressive attention into a dictionary lookup: binarized queries retrieve the value from the latest earlier matching key, or zero when none exists. Repeated keys overwrite, and the dictionary grows with distinct keys rather than context length. A two-layer, 64-dimensional model trained on eight associations reached 99.58%–99.73% recall at 4,096 on a controlled task, not language-model generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.