There will be a TensorBoard integration at some point in the future, which will enable interactive graphs.
For the moment, you can use this script, and an example of how to use it can be seen in this ipython notebook
For the moment, we need to wait until the new autograd is merged into master (which should happen next week). Once that’s done, one can start working on a tensorboard integration, as the underlying graph representation from PyTorch shouldn’t change much
Do we have any plans to build own visualization tool based on jit.trace or ONNX? I and my colleagues had built one basic visualizer and we were planning to make a better one. If somebody is working on it already, would like to discuss.