Hello. Though it is a little bit weird, I am wondering if saving the torch tensors (not the model weights) as numpy’s npz files will have any potential problems? What about the map_location=None setting if the torch tensor saved is on GPU?
I’ve tried to save it in npz file and load with np.load and it seems the tensor is correct (the same if I save with torch.save), even though I don’t even import torch when I np.load the saved tensor. Thank you so much in advanced.