MultiGPU training

Basically I have 2 gpus on hand and I want to you both of them for training sadly im not able to do so,
I followed the opennmt docs and used the parameters --worldsize 2 --gpu_ranks 0 1 but it doesnt work; it then wants to use the cpu for training.

1 Like

What command are you running to train? omnt_train?

You can try setting CUDA_VISIBLE_DEVICES

export CUDA_VISIBLE_DEVICES=0,1