Faculty Candidate Seminar

Hardware-Aware Efficient Primitives for Machine Learning

Dan FuPh.D. CandidateStanford University
3725 Beyster BuildingMap

Zoom link for remote participants: password 123123


Abstract: Efficiency is increasingly tied to quality to machine learning, with more efficient training algorithms leading to more powerful models. However, today’s most popular machine learning models are built on asymptotically inefficient primitives. For example, attention in Transformers scales quadratically in the input size, while MLPs scale quadratically in model dimension. In this talk, I discuss my work on improving the efficiency of the core primitives in machine learning, with an emphasis on hardware-aware algorithms and long-context applications. First, I focus on replacing attention with gated state space models (SSMs) and convolutions, which scale sub-quadratically in context length. I describe the H3 (Hungry Hungry Hippos) architecture, a gated SSM architecture that matches Transformers in quality up to 3B parameters and achieves 2.4x faster inference. Second, I focus on developing hardware-aware algorithms for SSMs and convolutions. I describe FlashFFTConv, a fast algorithm for computing SSMs and convolutions on GPU by optimizing the Fast Fourier Transform (FFT). FlashFFTConv yields up to 7x speedup and 5x memory savings, even over vendor solutions from Nvidia. Third, I will briefly touch on how these same techniques can also be used to develop sub-quadratic scaling in the model dimension. I will describe Monarch Mixer, which uses a generalization of the FFT to achieve sub-quadratic scaling in both sequence length and model dimension. Throughout the talk, I will give examples of how these ideas are beginning to take hold, with gated SSMs and their variants now leading to state-of-the-art performance in long-context language models, embedding models, and DNA foundation models.

Bio: Dan Fu is a Ph.D. student in the Computer Science Department at Stanford University, where he is co-advised by Christopher Ré and Kayvon Fatahalian. His research interests are at the intersection of systems and machine learning. Recently, he has focused on developing algorithms and architectures to make machine learning more efficient, especially for enabling longer-context applications. His research has appeared as oral and spotlight presentations at NeurIPS, ICML, and ICLR, and he has received the best student paper runner up at UAI. Dan has also been supported by an NDSEG fellowship.


Cindy Estell

Student Host

Sachin Garg

Faculty Host

Michal Derezinski