"checkpoint_model: true" not working

I can’t seem to get model checkpointing working. Here’s the relevant bit from my config.yml file:

  - name: DIETClassifier
    epochs: 100
    evaluate_on_number_of_examples: 50
    evaluate_every_number_of_epochs: 1
    checkpoint_model: true
    constrain_similarities: true

I’ve tried running training for 5 epochs. I expect it to save a new model checkpoint after every epoch (if the new checkpoint is better than the last). However, I don’t see it saving any model checkpoints in the models folder when I run rasa train or in the “checkpoints” folder when I run rasa train --out checkpoints. Any idea what I’m doing wrong?

What version of Rasa are you using?

I think, but I could be wrong, that the --out folder is meant for the final model. Per the docs:

| checkpoint_model                | False            | Save the best performing model during training. Models are   |
|                                 |                  | stored to the location specified by `--out`. Only the one    |
|                                 |                  | best model will be saved.                                    |
|                                 |                  | Requires `evaluate_on_number_of_examples > 0` and            |
|                                 |                  | `evaluate_every_number_of_epochs > 0`                        |

Rasa version: 2.8.6