Zhirui Feng If the model was built on CUDA, when you load on a CPU you use this:

torch.load(‘myfile.pth’, map_location=lambda storage, location: ‘cpu’)