A Metric Driven Approach to Mixed Precision Training
This work addresses efficiency challenges for deep learning practitioners by offering an incremental approach to optimize hardware usage through better numeric type selection.
The paper tackles the problem of high memory and compute costs in deep learning by proposing a metric-driven methodology for selecting low-precision numerics, demonstrating its application to scale training of a language representation model with potential generalization to other architectures.
As deep learning methodologies have developed, it has been generally agreed that increasing neural network size improves model quality. However, this is at the expense of memory and compute requirements, which also need to be increased. Various efficiency techniques have been proposed to rein in hardware costs, one being the use of low precision numerics. Recent accelerators have introduced several different 8-bit data types to help accommodate DNNs in terms of numerics. In this paper, we identify a metric driven methodology to aid in the choice of numerics. We demonstrate how such a methodology can help scale training of a language representation model. The technique can be generalized to other model architectures.