It would be nice to separate the tensorboard viewer and the process spawning, as some workflows prefer handling the spawning of the tensorboard process. It also makes it easier to implement logic for having seperate viewer address & tensorboard host binding address
It would be nice to separate the tensorboard viewer and the process spawning, as some workflows prefer handling the spawning of the tensorboard process. It also makes it easier to implement logic for having seperate viewer address & tensorboard host binding address