Describe the bug
Even after running nvflare poc prepare -d [image] -n 2 followed by nvflare poc start -gpu 0, the spawned containers do not have GPU capabilities enabled upon startup.
I am currently using a workaround by setting export GPU2USE='--gpus=0' manually before starting the POC.
To Reproduce
Run: nvflare poc prepare -d -n 2
Run: nvflare poc start -gpu 0
Check for GPU availability inside the container (e.g., check if PyTorch detects CUDA or run nvidia-smi).
Result: GPU is not detected unless the environment variable is manually exported.
Expected behavior
The -gpu flag in the nvflare poc start command should automatically configure the containers with the appropriate GPU capabilities without requiring manual environment variable exports.
Desktop:
- OS: ubuntu 22.04
- Python Version 3.10.12
- NVFlare Version 2.7.2
Describe the bug
Even after running nvflare poc prepare -d [image] -n 2 followed by nvflare poc start -gpu 0, the spawned containers do not have GPU capabilities enabled upon startup.
I am currently using a workaround by setting export GPU2USE='--gpus=0' manually before starting the POC.
To Reproduce
Run: nvflare poc prepare -d -n 2
Run: nvflare poc start -gpu 0
Check for GPU availability inside the container (e.g., check if PyTorch detects CUDA or run nvidia-smi).
Result: GPU is not detected unless the environment variable is manually exported.
Expected behavior
The -gpu flag in the nvflare poc start command should automatically configure the containers with the appropriate GPU capabilities without requiring manual environment variable exports.
Desktop: