Can't request multiple GPUs when deploying on Runpod

[alxiang]

in sky/provision/runpod/utils.py, gpu_quantity only factors into min_vcpu_count and min_memory_in_gb, so when I have my resources like:

resources:
  cloud: runpod
  accelerators: A100-80GB-SXM:2

I only get 1 A100 in Runpod and not 2. Based on my understanding of Runpod's Python SDK, the proper way to specify the # of GPUs is the gpu_count parameter (see https://github.com/runpod/runpod-python/blob/9e11c994fcceba0d6c3cd35be87084a28d9426d3/runpod/api/ctl_commands.py#L119C12-L119C21).

I'm on runpod==1.6.0, seems like skypilot should be updated to be compatible?

[Michaelvll]

That is a great catch @alxiang! Thank you for reporting this! Would you like to submit a PR to fix this?