Disable log versioning

[marcmk6]

I'm training a model and monitoring its convergence while training.
Say I first train it for 40 epochs, and I want to continue the training for another 40 epochs.
What I got in tensorboard logs are two separate folders version_0 and version_1, which makes the metrics two separate plots, instead of a continuous, single one.
I'm wondering if it's possible to disable this versioning stuff, so that the metrics logged during new training stages are showed together with the old ones.

The following image shows that the curve is split into two parts due to the versioning.

Thanks!

[marcmk6]

Well I think maybe this can be done by specifying the version parameter