LevAttention: Time, Space, and Streaming Efficient Algorithm for Heavy Attentions

A central problem related to transformers can be stated as follows: given two n × d matrices Q and K, and a non-negative function f, define the matrix A as follows: (1) apply the function f to each entry of the n × n matrix Q KT, and then (2) normalize each of the row sums of A to be equal to 1. The matrix A can be computed in O(n2 d) time assuming f can be applied to a number in constant time, but the quadratic dependence on n is prohibitive in applications where it corresponds to long context lengths. For a large class of functions f, we show how to find all the ``large attention scores", i.e., entries of A which are at least a positive value varepsilon, in time with linear dependence on n (i.e., n · textrmpoly(d/varepsilon)) for a positive parameter varepsilon > 0. Our class of functions include all functions f of the form f(x) = |x|p, as explored recently in transformer models. Using recently developed tools from randomized numerical linear algebra, we prove that for any K, there is a ``universal set" U subset [n] of size independent of n, such that for any Q and any row i, the large attention scores Ai,j in row i of A all have j ∈ U. We also find U in n · textrmpoly(d/varepsilon) time. Notably, we (1) make no assumptions on the data, (2) our workspace does not grow with n, and (3) our algorithms can be computed in streaming and parallel settings. We call the attention mechanism that uses only the subset of keys in the universal set as LevAttention since our algorithm to identify the universal set U is based on leverage scores. We empirically show the benefits of our scheme for vision transformers, showing how to train new models that use our universal set while training as well, showing that our model is able to consistently select ``important keys'' during training.