Flash Invariant Point Attention

Andrew Liu, Axel Elaldi, Nicholas T Franklin, Nathan Russell, Gurinder S. Atwal, Yih-En Andrew Ban, Olivia Viessmann

Keywords: geometric deep learning, structural biology, invariant point attention

TL;DR: We reduce IPA, one of the most popular geometric deep learning methods in structural biology, from quadratic to linear scaling without performance loss, and train on previously non-trainable data.

Abstract: 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.

Primary Area: Machine learning for sciences (e.g. climate, health, life sciences, physics, social sciences)

Submission Number: 10822