How to add parameter to select GPU at runtime?

[setuc]

Hi,
how do we add a command line parameter to the select the GPU at run time, especially on amulti gpu machine?
I tried adding

@action.command()

@click.option('-g', '--gpu', help='select gpu', default=1, required=true)
def select_gpu(gpu):
    os.environ["CUDA_VISIBLE_DEVICES"]=gpu

I tried to add a similar code block for train_evaluate_predict_pipeline and pass the gpu parameter
but i keep getting invalid option during runtime. I know its not the package issue, but the documentation did not help either

[jakubczakon]

I am working on multigpu machines and I usually do it just by going along these lines:

CUDA_VISIBLE_DEVICES=1 neptune run -- train_evaluate_predict_pipeline -p fasttext_gru

and for multigpu you just do:

CUDA_VISIBLE_DEVICES=1,2 neptune run -- train_evaluate_predict_pipeline -p fasttext_gru

I haven't tried multigpu for this project yet.

Also apart from the documentation we have quite a vibrant community here https://community.neptune.ml/
drop your issue there and I am sure someone will help.

[kamil-kaczmarek]

@setuc Also, you can do $ export CUDA_VISIBLE_DEVICES=1, to select particular GPU, for entire terminal session.