Rasa model not loading on Azure

Rasa Open Source
1

Good day,

I am working on creating a chatbot, with actions, tts, etc. That is then deployed as a web app. I have a fully working version on localhost without any issues.

However, when I try to deploy the nlu on Azure, either as docker compose or singe image. It always gets stuck with loading the model.

I tried every possible idea that my colleges or GPT had, but nothing is working and I dont get any further logs. Here are the logs I get:

2025-11-10T11:01:18.7562968Z 2025-11-10 11:01:18 DEBUG rasa.utils.tensorflow.models - Loading the model from /tmp/tmpo_m3mha8/train_DIETClassifier5/DIETClassifier.tf_model with finetune_mode=False…

2025-11-10T11:01:19.7369901Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - Following metrics will be logged during training:

2025-11-10T11:01:19.7370404Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - t_loss (total loss)

2025-11-10T11:01:19.7370524Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - i_acc (intent acc)

2025-11-10T11:01:19.7370571Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - i_loss (intent loss)

2025-11-10T11:01:19.7370611Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - e_f1 (entity f1)

2025-11-10T11:01:19.7370651Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - e_loss (entity loss)

2025-11-10T11:01:19.7370696Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - r_f1 (role f1)

2025-11-10T11:01:19.7370736Z 2025-11-10 11:01:19 DEBUG rasa.nlu.classifiers.diet_classifier - r_loss (role loss)

2025-11-10T11:01:28.4403374Z /usr/lib/python3.10/random.py:370: DeprecationWarning: non-integer arguments to randrange() have been deprecated since Python 3.10 and will be removed in a subsequent version

2025-11-10T11:01:28.4404042Z return self.randrange(a, b+1)

and then nothing.

I am currently using rasa version 3.6.21-full and even reduced the model size to 1MB.

If anyone could help me out, that would be really appreciated. I am getting rally desperate, have been sitting on this for 6 weeks now

2

Hey, Let’s break this down. The log line where it gets stuck is very telling:
Loading the model from /tmp/tmpo_m3mha8/train_DIETClassifier5/DIETClassifier.tf_model with finetune_mode=False…


The fact that it hangs here, especially after the “Following metrics will be logged during training” debug message, strongly points to an issue with the TensorFlow model loading process in your Azure environment.

It’s either Azure’s CPU doesn’t support the instruction sets TensorFlow expects or Insufficient memory.
In your Dockerfile, make sure you are copying the model into a fresh location and that the rasa run command is pointing to it correctly. A good practice is to use a dedicated directory like /app/models.