jaan is looking for a way to dynamically set the resources spec in SkyPilot to check availability zones for GPUs, as the regions change daily for what has capacity. Zongheng Yang suggested overshooting by setting multiple regions initially and letting SkyPilot handle out-of-capacity errors. jaan is interested in dynamically setting the resources spec.
jaan
Asked on Apr 06, 2024
You can dynamically set the resources spec in SkyPilot by writing a simple script.
One approach could be to overshoot by setting multiple regions initially and letting SkyPilot handle out-of-capacity errors.
Another option is to use the --retry-until-up / -r
flag with sky launch
to automatically retry all locations until a functional node is found.