All pipelines include a training step. Currently, all training is powered by Tesla K80.
TFRecords and speed up
TFRecords is a binary data format defined by the Tensorflow project. Its purpose is to make the reading of data in deep learning pipelines more efficient and therefore faster.
For the Core Engine, we automatically encode all data flowing through our pipelines in TFRecords format. We have optimized our training loops to take full advantage of the TFRecords format, which has been shown to increase preprocessing and training speed by up to 7 times, depending on the data and configuration.
As a user, you get this for all your pipelines, built-in and ready to go.