2023, Rasa 3.x and still the same issue!! Did you ever find a solution for this? Looking into cache folder, I also see that not more than a few checkpoints are saved. My validation plot is not complete… Please let me know if you know how to fix this issue 
Hello. Is it possible that in the current Rasa 3.5.6 not all of the individual losses get registered into Tensorboard?
The only loss I can see is epoch_t_loss which I understand is the total loss (the combination of the three losses of the DIET model).