Reproduce “Putting Transformer Theories to the Test” 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
Awni Altabaa, Tobias Schnabel, Kiran Tomlinson
The authors compare seven theories of transformer capabilities with small models and pretrained language models across several tasks. They distinguish expressivity from whether training finds solutions, and test softmax against uniform attention on query-dependent retrieval. At key dimension 6, softmax models reach 100% accuracy while uniform-attention models reach about 55%; this synthetic result does not identify why uniform attention fails.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.