Adaptive Loss Scaling for Mixed Precision Training

Mixed precision training (MPT) is becoming a practical technique to improve the speed and energy efficiency of training deep neural networks by leveraging the fast hardware support for IEEE half-precision floating point that is available in existing GPUs. MPT is typically used in combination with a technique called loss scaling, that works by scaling up the loss value up before the start of back-propagation in order to minimize the impact of numerical underflow on training. Unfortunately, existing methods make this loss scale value a hyperparameter that needs to be tuned per-model, and a single scale cannot be adapted to different layers at different training stages. We introduce a loss scaling-based training method called adaptive loss scaling that makes MPT easier and more practical to use, by removing the need to tune a model-specific loss scale hyperparameter. We achieve this by introducing layer-wise loss scale values which are automatically computed during training to deal with underflow more effectively than existing methods. Training deep neural networks (DNNs) is well-known to be time and energy consuming, motivating the development of new methods and hardware to make training more efficient. One way to improve training efficiency is to use numerical representations that are more hardware-friendly. This is the reason that the IEEE 754 32-bit single-precision floating point format (FP32) is more widely used for training DNNs than the more precise double precision format (FP64), which is commonly used in other areas of high-performance computing. In an effort to further improve hardware efficiency, there has been increasing interest in using data types with even lower precision than FP32 for training (Micikevicius et al., 2018; Kuchaiev et al., 2018; Wang et al., 2018; Kalamkar et al., 2019; Mellempudi et al., 2019; Sakr et al., 2019).