At the moment, passing GPUs to docker does not automatically set the CUDA_VISIBLE_DEVICES environment variable within the container. This variable is needed in some situations, and it has a direct relationship to the gpus that are being passed to the MLCube through --gpus. Automatically setting that variable would be a great feature for MLCube.
At the moment, passing GPUs to docker does not automatically set the
CUDA_VISIBLE_DEVICESenvironment variable within the container. This variable is needed in some situations, and it has a direct relationship to the gpus that are being passed to the MLCube through--gpus. Automatically setting that variable would be a great feature for MLCube.