search

shindavid / AlphaZeroArcade

6 stars 1 forks source link

Running on the cloud #61

Open shindavid opened 1 year ago

shindavid commented 1 year ago

We want to be able to run the AlphaZero main loop, along with related processes (e.g., see https://github.com/shindavid/AlphaZeroArcade/issues/59), on the cloud. AWS is the obvious choice. Crusoe Cloud is a competitor that undercuts AWS on price - I am good friends with the CEO.

I have no experience with cloud compute solutions, so I don't even know where to start. I am willing to create an organization and bill everything to a credit card of mine. But I would want some sorts of protections in place to make sure an accidental rogue experiment launched by someone (whether myself or a different project member) doesn't result in a massive bill.

This requires at a minimum an environment setup mechanism (https://github.com/shindavid/AlphaZeroArcade/issues/60).

We likely want custom dashboards to track experiments, visualize results, etc.

xuanchenghuang commented 1 year ago

There are some free resources we could initially use for cloud development setups (in case someone doesn't have a GPU for example).

Right now what I'm testing setup on is https://aws.amazon.com/sagemaker/studio-lab/ which is free to sign up for and use (max 12 hours/session for CPU, 4 hours/session for GPU). There are other free utils such as Google Colab, but a lot of them don't support Docker so it would be difficult to develop non-Python apps/scripts.