I’ve always used torch.no_grad() and recently discovered the torch.inference_mode. Inference_mode is the prefered way for most of the users but when I tested it seems that is a little slower then torch.no_grad(). Does anyone had this impression too?
I’ve always used torch.no_grad() and recently discovered the torch.inference_mode. Inference_mode is the prefered way for most of the users but when I tested it seems that is a little slower then torch.no_grad(). Does anyone had this impression too?