Using rocAL with TensorFlow for training

rocAL improves machine learning (ML) pipeline efficiency by preprocessing data and parallelizing data loading.

TensorFlow iterators and readers are provided as plugins to separate data loading from training.

You’ll need a rocAL TensorFlow Docker container to run TensorFlow training with rocAL.

To use rocAL with TernsorFlow, import the rocAL TensorFlor plugin:

from amd.rocal.plugin.tf import ROCALIterator

Set up a training pipeline that reads data with readers.tfrecord and uses decoders.image to decode the raw images.

Call the training pipeline using ROCALIterator.

An example of TensorFlow training using rocAL is available in the rocAL GitHub repository.