I\'m currently working on some computer vision models in both PyTorch and Tensorflow. Weirdly, I\'ve run into a huge performance gap. I\'ve implemented AlexNet as equivalent