Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation
This work addresses the need for efficient deep learning inference in resource-constrained environments, presenting an incremental improvement with a practical workflow for 8-bit quantization.
The paper tackles the problem of reducing deep neural network size and improving inference performance through integer quantization, achieving within 1% accuracy of floating-point baselines across various models including MobileNets and BERT-large.
Quantization techniques can reduce the size of Deep Neural Networks and improve inference latency and throughput by taking advantage of high throughput integer instructions. In this paper we review the mathematical aspects of quantization parameters and evaluate their choices on a wide range of neural network models for different application domains, including vision, speech, and language. We focus on quantization techniques that are amenable to acceleration by processors with high-throughput integer math pipelines. We also present a workflow for 8-bit quantization that is able to maintain accuracy within 1% of the floating-point baseline on all networks studied, including models that are more difficult to quantize, such as MobileNets and BERT-large.