Training in Rasa-X fails with EmbeddingPolicy configuration

I have the following policy configured:

- name: EmbeddingPolicy
  epochs: 1000
  attn_shift_range: 3
  - name: FullDialogueTrackerFeaturizer
    - name: LabelTokenizerSingleStateFeaturizer

If I train using Rasa-X I get the following error: - 500, Internal Server Error, body='b'{"version":"1.6.2","status":"failure","message":"An unexpected error occurred during training. Error: Found two unpredictable actions in one story.Check your story files.","reason":"TrainingError","details":{},"help":null,"code":500}''

But if I train in the CLI using rasa train --augmentation 0 everything works fine.

Is there a way to have Rasa-X use the augmentation option when training?

I have filed an enhancement issue for this, in the meantime you can train locally and use the upload option in Rasa X to upload the model.

1 Like