I have 2x GPU's and the script identifies both of them and reports "Model pushed to 2 GPU(s)" , type geforce etc. However, only 1x GPU is being used. At the very least, I think you need to wrap the model in the nn.DataParallel wrapper and add an autocast.
I have 2x GPU's and the script identifies both of them and reports "Model pushed to 2 GPU(s)" , type geforce etc. However, only 1x GPU is being used. At the very least, I think you need to wrap the model in the nn.DataParallel wrapper and add an autocast.