MLEM helps you package and deploy machine learning models. It saves ML models in a standard format that can be used in a variety of production scenarios such as real-time REST serving or batch processing.
-
Run your ML models anywhere: Wrap models as a Python package or Docker Image, or deploy them to Heroku, SageMaker or Kubernetes (more platforms coming soon). Switch between platforms transparently, with a single command.
-
Model metadata into YAML automatically: Automatically include Python requirements and input data needs into a human-readable, deployment-ready format. Use the same metafile on any ML framework.
-
Stick to your training workflow: MLEM doesn't ask you to rewrite model training code. Add just two lines around your Python code: one to import the library and one to save the model.
-
Developer-first experience: Use the CLI when you feel like DevOps, or the API if you feel like a developer.
The main reason to use MLEM instead of other tools is to adopt a GitOps approach to manage model lifecycles.
-
Git as a single source of truth: MLEM writes model metadata to a plain text file that can be versioned in Git along with code. This enables GitFlow and other software engineering best practices.
-
Unify model and software deployment: Release models using the same processes used for software updates (branching, pull requests, etc.).
-
Reuse existing Git infrastructure: Use familiar hosting like Github or Gitlab for model management, instead of having separate services.
-
UNIX philosophy: MLEM is a modular tool that solves one problem very well. It integrates well into a larger toolset from Iterative.ai, such as DVC and CML.
This a quick walkthrough showcasing deployment functionality of MLEM.
Please read Get Started guide for a full version.
MLEM requires Python 3.
To install the pre-release version:
Check out what we have:
Click to show `cat` outputIf you want to follow this Quick Start, you'll need to sign up on https://heroku.com, create an API_KEY and populate HEROKU_API_KEY env var (or run heroku login in command line). Besides, you'll need to run heroku container:login. This will log you in to Heroku container registry.
Now we can deploy the model with mlem deploy (you need to use different app_name, since it's going to be published on https://herokuapp.com):
Contributions are welcome! Please see our Contributing Guide for more details.
Thanks to all our contributors!
This project is distributed under the Apache license version 2.0 (see the LICENSE file in the project root).
By submitting a pull request to this project, you agree to license your contribution under the Apache license version 2.0 to this project.
