To run TensorFlow on a remote machine, you will need to manage it yourself—installing the correct software, making sure it is up to date, and turning the server on and off. While it is still possible to do so for one machine, and you sometimes need to distribute the training among numerous GPUs, using Google Cloud ML to run TensorFlow allows you to focus on your model and not on operations.
You will find that Google Cloud ML is useful for the following:
- Training your model quickly thanks to elastic resources in the cloud
- Looking for the best model parameters in the shortest amount of time possible using parallelization
- Once your model is ready, serving predictions without having to run your own prediction server
All the details for packaging, sending, and running your model are available in the Google Cloud ML documentation (https://cloud.google.com/ml-engine/docs/).
