Reproduce “FlashCart: Fast Cartesian Tensor Products for Equivariant Interatomic Potentials” 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
Viktor Zaverkin, Payman Goodarzi, Sergey V. Sukhomlinov, Davit Hovhannisyan, Roland Aydin, Martin H. Müser, Mathias Niepert
FlashCart combines generated GPU kernels with a recursive feature ladder that compresses features to fixed channel width after each product. In tested configurations, higher correlation order improved accuracy more efficiently than changes to width, depth, or rank. On SPICE-MACE-OFF, its 5.6-million-parameter model had lower energy and force MAEs than 188.9-million-parameter PET-L and ran about 10× faster; this result is benchmark- and setup-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.