Automl Model Deploy Fails On Gunicorn Version Conflict

gwinner 0

Hello,

I am attempting to deploy an automl model to a real time endpoint, when I do I am getting a pip error about dependency conflicts between gunicorn and mlflow. It seems these install are dynamically added by azure? What can I do to get past this error?

Thanks! AutoML Endpoint Logs.txt

AshokPeddakotla-MSFT 35,581 Reputation points

2025-01-06T03:25:20.0766667+00:00

gwinner Greetings & Welcome to Microsoft Q&A forum!

As error suggests that there is a conflict between the versions of gunicorn required by mlflow and azureml-inference-server-http. Specifically, mlflow requires gunicorn to be less than version 22, while azureml-inference-server-http requires a version of gunicorn that is 22 or greater.

Are you following any documentation? Did you check How to deploy an AutoML model to an online endpoint for the steps and pre-requisites?

Also, review your conda.yaml file and ensure that the dependencies are compatible with each other.
gwinner 0 Reputation points

2025-01-06T13:24:24.7666667+00:00
Hi thank you for the help,

As I am learning and doing this in azure for the first time, yes I did follow the deploy docs, the conda file was auto generated by Automl when the model was created. The generated Automl model includes a conda file that specifies the mlflow version, but not gunicorn nor the inference-server-http package.

Hence my quesion, I have no control over two of these packages, they are added when I choose to deploy an endpoint. The mlflow pacakge I could update, but I am not sure if that affect the model that automl created or if their is a way to update this not just locally by also in the model inside of azure.

Conda file contents:

channels:

anaconda

conda-forge

pytorch

dependencies:

python=3.9.20

pip:

mlflow==2.9.2

azureml-train-automl-runtime==1.58.0

inference-schema

xgboost<=1.5.2

prophet==1.1.4

azureml-interpret==1.58.0

azureml-defaults==1.58.0

pytorch-transformers==1.0.0

spacy==3.7.4

https://aka.ms/automl-resources/packages/en_core_web_sm-3.7.1.tar.gz

numpy==1.23.5

pandas==1.5.3

scikit-learn==1.5.1

holidays==0.62

psutil==5.9.3

pytorch==2.2.2

cudatoolkit==9.0

name: project_environment

Share via

Automl Model Deploy Fails On Gunicorn Version Conflict

Your answer