While being able to see into our training process and comparing models is obviously pretty great, that's not all TensorBoard can do. We can also use it to visualize network structure. Here, I've navigated to GRAPHS and pulled up the structure of the deep neural network:

The training node represents the input tensor, and by default, it's this giant octopus that connects to the rest of the graph in a somewhat unhelpful way. To fix that, you can just click the node and click Remove from the main graph. It will then be moved off to the side.