Using a cloud computing provider is useful to offload computing to faster machines. It can also be used if we want to make multiple runs at the same time. For example, we could try fixing exploding gradients by launching two runs: one with a lower learning rate and one with a lower gradient clipping. We could spawn two different VMs, each training its own model, and see which performs better.
We are going to use Google Cloud Platform (GCP), but other cloud providers, such as Amazon AWS or Microsoft Azure, will also work. We'll go through the different steps needed to train a Melody RNN model on the piano jazz dataset from the previous chapter, including the GCP account configuration and VM instance creation.
