Large Memory Layers with Product Keys

TL;DR

This paper presents a large, efficient memory layer based on product keys that significantly enhances neural network capacity, enabling better performance on large-scale language modeling tasks with minimal computational overhead.

Contribution

Introduction of a structured memory layer using product keys that allows for extremely large capacity and fast nearest neighbor search, improving language modeling efficiency.

Findings

Memory-augmented model outperforms deeper baseline with fewer layers.

Memory layer enables training on datasets with up to 30 billion words.

Model with memory is twice as fast at inference while achieving better accuracy.

Abstract

This paper introduces a structured memory which can be easily integrated into a neural network. The memory is very large by design and significantly increases the capacity of the architecture, by up to a billion parameters with a negligible computational overhead. Its design and access pattern is based on product keys, which enable fast and exact nearest neighbor search. The ability to increase the number of parameters while keeping the same computational budget lets the overall system strike a better trade-off between prediction accuracy and computation efficiency both at training and test time. This memory layer allows us to tackle very large scale language modeling tasks. In our experiments we consider a dataset with up to 30 billion words, and we plug our memory layer in a state-of-the-art transformer-based architecture. In particular, we found that a memory augmented model with only 12 layers outperforms a baseline transformer model with 24 layers, while being twice faster at inference time. We release our code for reproducibility purposes.