First just follow the instructions on github on how to run the crayon server.
Then, just initialize your model regularly and add the pycrayon attributes to it:
self.cc = CrayonClient(hostname="localhost")
# Create a new experiment
self.summary = self.cc.create_experiment(type(self).__name__)
Then, when you compute you training loss, for example, you can do:
You can do the same process to any scalar value you are interested in monitoring. Crayon currently only supports scalars and histograms.
To see what's going on on TensorBoard, just type in your browser: