The weights that are optimized are always leaves of the computational graph, so we only need to care about them.SummarizingWhenever you are working with PyTorch, the steps you should follow are these:Define your Network class by placing the layers with weights that can be updated inside the __init__ method..Then define how the data flows through the layers inside the forward method.Define how your data should be loaded using the Dataset class..Then use DataLoader class to loop over your data.Choose an optimizer and a loss function..Loop over your training data and let the optimizer update the weights of your network.Creating good models might take a lot of practice and experience..With this in mind there is an excellent library based on PyTorch that already implements lots of things to you..I would highly recommend looking into it..It is called Fastai.After you train your model, the next step is putting it somewhere people can use it..If you want to know a little more about putting your model into production, I created an article that explains it more..You can access it here.Let me know what you thought about the post..Check me out on any of these places:pedro-torres.comLinkedInTwitter. More details