We need to be able to set a specific GPU for training. This is usually done by setting the devices parameter of the trainer to devices=[2] for the second GPU. As the devices gets used to get the number of GPUs, this parameter is expected to be an int.
As a workaround you can set the GPU in the devcontainer.json:
"runArgs": [
"--gpus",
"device=2",
...
But you need to rebuild the container and reinstall the dependencies.
We need to be able to set a specific GPU for training. This is usually done by setting the
devices
parameter of thetrainer
todevices=[2]
for the second GPU. As thedevices
gets used to get the number of GPUs, this parameter is expected to be an int.As a workaround you can set the GPU in the
devcontainer.json
:But you need to rebuild the container and reinstall the dependencies.