narenst/infinity

Add support for spot instances

narenst opened this issue · 1 comments

Infinity should allow the users to spin up spot instances. Spot instances are much cheaper than on-demand instances. But they may be preempted in the middle of a training. In that case, infinity should let the user recover from the failure and create a new spot instance.

If anyone looking at this ticket has more requirements to add, please do so.