5fa1a76
1
Users should specify the torch_dtype they want, and if they don't it will be torch.float32.
