hparams with separate train and validation summary writers

I have a model that I train with hp.KerasCallback and TensorBoard callback as explained in the example:

model.fit(
    ...,
    callbacks=[
        tf.keras.callbacks.TensorBoard(logdir),  # log metrics
        hp.KerasCallback(logdir_hp, hparams),  # log hparams
    ],
)

This successfully prints all my input (hyper)parameters (in hparams dict) and

• train metrics, if logdir_hp is /train (or a summary writer to train dir)
• validation metrics, if logdir_hp is /validation (or a summary writer to validation dir)

but never both because of the ambiguous summary tensor name!

I expect to log the same metric but in both runs (train and validation) using prefixes (like acc and val_acc) for better analysis

The above result was obtained using estimator api which automatically appends the run name based on the folder (train on current folder, so no prefix, for validation it's eval and I had an additional evaluation every x steps saved in global_eval), it would be great to achieve the same in keras model.fit()