Mixed Precision Training With 8-bit Floating Point

TL;DR

This paper introduces a novel method for training deep neural networks using 8-bit floating point precision, achieving high accuracy and efficiency improvements across multiple datasets and models.

Contribution

The paper presents a new approach for 8-bit floating point training, including loss scaling and stochastic rounding, enabling state-of-the-art accuracy with reduced precision.

Findings

Achieved state-of-the-art accuracy on ImageNet and WMT16 datasets.

Demonstrated effective training of various models like ResNet and Transformer at 8-bit precision.

Reported slightly higher validation accuracy than full precision baseline.

Abstract

Reduced precision computation for deep neural networks is one of the key areas addressing the widening compute gap driven by an exponential growth in model size. In recent years, deep learning training has largely migrated to 16-bit precision, with significant gains in performance and energy efficiency. However, attempts to train DNNs at 8-bit precision have met with significant challenges because of the higher precision and dynamic range requirements of back-propagation. In this paper, we propose a method to train deep neural networks using 8-bit floating point representation for weights, activations, errors, and gradients. In addition to reducing compute precision, we also reduced the precision requirements for the master copy of weights from 32-bit to 16-bit. We demonstrate state-of-the-art accuracy across multiple data sets (imagenet-1K, WMT16) and a broader set of workloads (Resnet-18/34/50, GNMT, Transformer) than previously reported. We propose an enhanced loss scaling method to augment the reduced subnormal range of 8-bit floating point for improved error propagation. We also examine the impact of quantization noise on generalization and propose a stochastic rounding technique to address gradient noise. As a result of applying all these techniques, we report slightly higher validation accuracy compared to full precision baseline.