@adeottii on Instagram: "So, Neural network-based systems are horribly hard to debug but you can actually just print and inspect the computational graph generated by your entire code…I write PyTorch code and I use TorchViz to print the computational graph (not sure what the Tensorflow equivalent is) That trace just shows all the tensors in my code that have gradients enabled. You sometimes don’t want gradients to flow through every single tensor in your code ( that’s why we use .detach() or torc