Question: any workaround for profiling on GPU-renting services?

[ArthurConmy]

Currently the best GPU rental sites I know of are vast.ai/runpod (LambdaLabs rarely has availability of good GPUs)

Both of these services use Docker containers, allow custom docker containers, but don't support --cap-add sys_ptrace.

For example, this 2022 message from vast admins clarifies that the ADD_PTRACE option is not allowed:

Given the constraints of what runpod and vast allow, is there any workaround for using py-spy with their GPUs? Thanks!