Remove gpu_id from saved fnet model state

[counkomol]

[Gamalol]

Hello Chek,

My name is Gamaliel, thank you very much for providing the label-free prediction method on github!, we are currently reproducing the results using the database you kindly provided, we have managed to perform several trainings using that dataset, but we are facing a problem regarding the use of GPUS, I quote the stated line on your manual:

Now to train your model on your dataset you would run (assuming you only have 1 GPU on slot 0)

./scripts/train_mymodel.sh mydata 0

We have the opportunity to use a 4 GPU Rig to perform the training but we have tried changing the '0' parameter to other numbers (1,2,3,4), removing the parameter, and everytime the system ends up using only 1 GPU. We have also checked other scripts to see if we can find the definition of GPU number usage but unfortunately we haven't succeeded on that. Could you please give me some guidance on this? We wish to change that parameter if possible to make use of 4 GPU. Thank you very much in advance.

Best regards, Gamaliel.

[counkomol]

Hi Gamaliel,

You might want to try editing your train_mymodel.sh so you can specify more than one GPU ID. I'm not sure though if the version of the code you are working with is capable of multi-GPU training without additional modifications.

Chek