Assuming there are 4 GPUs, how do I train on the GPU of choice instead of all the GPUs? I tried fiddling around with n_devices and device_id variables in main.py, but in vain.
@Meghana-Meghana you can set environment variable CUDA_VISIBLE_DEVICES, e.g. CUDA_VISIBLE_DEVICES=0 gpt-2 ... or to use two GPUs CUDA_VISIBLE_DEVICES=0,1 gpt-2 ...
Assuming there are 4 GPUs, how do I train on the GPU of choice instead of all the GPUs? I tried fiddling around with n_devices and device_id variables in main.py, but in vain.