Syncing the epoch numbers of Model.fit and TensorBoard

I would like to discuss https://github.com/tensorflow/tensorboard/issues/5805 here:

keras.Model.fit prints epoch numbers, starting with 1. However, using keras.callbacks.TensorBoard yields TensorBoard log curves that start at a Step value of 0 (and the steps correspond to epochs, see TB#5806).

Stanley Bileschi suggests that this might be fixable by Keras passing an explicit step value, starting with 1, so I’m asking any Keras developers around: Would this work? Would this be good to have?

Hi Robert! Yes. you can explicitly set “epoch or steps value = 1” to see changes in Tensorboard

@Manas_Mohanty So how and where could I set this? In the TensorBoard callback’s code?