Try setting your model to evalaution mode with my.eval() before writing it to tensorboard.

This reduces the randomness in your model for each pass. As the add_graph method calls the graph several times for tracing, it errors when differences happend due to the random nature of the model.