Right now, I am getting an error when I try launching rasa open-source with docker-compose up. The error(s) that I get include the following: /opt/venv/lib/python3.7/site-packages/rasa/shared/utils/io.py:93: UserWarning: The model at 'model-location-here' could not be loaded. Error: <class 'ValueError'>: Wrong loss type 'cross_entropy', should be 'margin' or 'softmax' and /opt/venv/lib/python3.7/site-packages/rasa/shared/utils/io.py:93: UserWarning: Agent could not be loaded with the provided configuration. Load default agent without any model.
Has anybody encountered this error(s) before?
If so, what was the solution? I have a feeling that this is due to the DIET Classifier potentially, but I could be wrong. When testing this model outside of the instance that houses docker, the most recent model works no problem.
Hi @jeff-ridgeway and thanks for your question. Could you share the version of rasa you’re using? On the latest version, cross_entropy should work just fine.
Hey @m-vdb, I just checked the docker version and right now I’m running rasa/rasa:2.2.2-full. So I updated that version to the current rasa/rasa:2.5.1-full and re-ran my docker image. Still getting the same error as stated originally. Thanks for the hint about the version! Will I always have to update the rasa version docker image in my DockerFile and do you have any other thoughts about how to proceed with that error?
Hey @m-vdb I was able to get the solution. It was basically that I was running the newest version of Rasa on Sagemaker but when I had to port over to my AWS ec2 instance, I had to make sure that was running the newest version of in my docker image, which is apart of my docker compose. Also had to make sure I was building the docker image properly with docker build. Thanks for pointing me in that direction and helping me. For anyone running into that issue, check out this GitHub resolved issue Is there a recommended fix for error Type with the DIET Classifier? · Issue #8609 · RasaHQ/rasa · GitHub