Mixed Precision Training — Less RAM, More Speed

Optimisation

Mixed Precision Training — Less RAM, More Speed

Speed up your models with two lines of code

When it comes to large complicated models it is essential to reduce the model training time as much as possible, and utilise the available hardware efficiently. Even small gains per batch or epoch are very important.

Mixed precision training can both significantly reduce GPU RAM utilisation, as well as speeding up the training process itself, all without any loss …

16-bit machine learning mixed mixed-precision neural networks optimisation precision training