How to deploy mlflow models to sagemaker endpoints where sagemaker refers the private docker registry?

[Maverick1]

Is it possible to deploy the mlflow model to a sagemaker endpoint where the image URL is not referring to an image in ECR but the image is actually present in a private docker registry?

[Kaniz]

Hi @Saurabh Verma​ , Can you pull the image from the private docker registry using kubernetes?

[Maverick1]

@Kaniz Fatma​ : Yes. Even when I am deploying the model using sagemaker studio, I am able to pull the image from private docker registry because the VPC settings for image and containers is allowed via Sagemaker API. But I can't see the same options available for mlflow-sagemaker API. The only option is VPC settings for the region where model need to be deployed, but in the end it needs the access to ECR repo of the same region.

[Atanu]

@Saurabh Verma​ , this to create the endpoint.

also, check this out - https://github.com/mlflow/mlflow/blob/0fa849ad75e5733bf76cc14a4455657c5c32f107/mlflow/sagemaker/__in...

[Kaniz]

Hi @Saurabh Verma​ , Did @Atanu Sarkar​ 's response help answer your question?

[Maverick1]

@Kaniz Fatma​ : Unfortunately No. The above implementation always assumes that image is hosted in ECR repository.