job-deploy-app

Deploy ML models from MLFlow model registry to Neu.ro platform job without line of code.

Supported inference servers:

• MLFlow
• Triton

Usage

If you want to run this application by your own, clone this repo and:

1. Build application container image:
   • neuro-flow build app
2. Run application:
   • neuro-flow run app

TODOs:

• Expose proper Triton model endpoint URIs
• Add ability to remove model deployments
• Monitor model endpoint health
• Support deployments to Seldon with mlflow2seldon
• Add documentation
  • pitfalls with deployment to mlflow:
    • container image should contain conda available via Path
    • container image should contain mlflow[extras]>=1.25 installed in it
  • pitfalls with deployment to mlflow:
    • onnx only ML models are supported
    • container image should be built on the compatible CUDA drivers