Setup GCP Compute Engine

[achen353]

Some of the language models are going to be humongous. Need to set up a Compute Engine instance on GCP (probably multi-GPU)

Useful Reference/Tutuorial

• How to run Deep Learning models on Google Cloud Platform in 6 steps?
• Setting up Jupyter Lab Instance on Google Cloud Platform
• Installing CUDA on Google Cloud Platform in 10 minutes
• Setting up Chrome Remote Desktop for Linux on Compute Engine

Hopefully, the tutorials in these links are still valid (used them more than a year ago). If anything does not apply or there exists better tutorials, please let the others know.

TODO

• Request GPU quotas
• Set up Compute Engine VM instance

[achen353]

Requested:

• K80 * 1
• V100 * 1
• P100 * 1

[achen353]

One annoying thing about GCP is that you need to request GPU quotas before you create a VM instance with GPU capabilities (https://cloud.google.com/compute/quotas). I requested three types (one for each) but based on my experience they won't approve all of them and depending on the availabilities it could take a while to get approved.

@stephanieeechang @andywang268 I highly suggest you also activate your own personal GCP account and request the GPU quotas for your account to have a higher chance of getting GPUs for either your personal accounts or our project account.