Invariant Point Attention (IPA) is a key algorithm for geometry-aware modeling in structural biology, central to many protein and RNA models. However, its quadratic complexity limits the input sequence length. We introduce FlashIPA, a factorized reformulation of IPA that leverages hardware-efficient FlashAttention to achieve linear scaling in GPU memory and wall-clock time with sequence length. FlashIPA matches or exceeds standard IPA performance while substantially reducing computational costs. FlashIPA extends training to previously unattainable lengths, and we demonstrate this by re-training generative models without length restrictions and generating structures of thousands of residues. FlashIPA is available at https://github.com/flagshippioneering/flash_ipa.
@article{arxiv.2505.11580,
title = {Flash Invariant Point Attention},
author = {Andrew Liu and Axel Elaldi and Nicholas T Franklin and Nathan Russell and Gurinder S Atwal and Yih-En A Ban and Olivia Viessmann},
journal= {arXiv preprint arXiv:2505.11580},
year = {2025}
}