Return the ID of a GPU which has no compute processes running on it.
GPUs which have compute processes running on it according to NVML, and GPUs for
which a different instance of getgpu has returned as available in the last
--startup-time seconds (30 by default), are treated as occupied.
To install:
pip install --user git+https://github.com/rshin/getgpuTo set CUDA_VISIBLE_DEVICES:
getgpu -- [program to run] ...Run getgpu --help to see options.
pip install -e . dev
pre-commit install