Thanks to ODS MLOps course I developed MLOps template project which consists of three repositories:
Those repositories can be used as template for new ML model projects development.
Model train
Repository: https://gitlab.com/alexakimenko/mlops_voice_train
Key features:
- MLFlow server for tracking experiments
- Gitlab CI for basic checks
- DVC for creation and running pipelines
- Poetry to lock dependencies
- Cookiecutter structure
Model serving
Repository: https://gitlab.com/alexakimenko/mlops_voice_serving
Key features:
- MLFlow server model registry for loading models
- FastAPI for http server
- Gitlab CI for CI/CD
- Docker for application delivery
- Prometheus for app monitoring
MLOps infrastructure
Repository: https://gitlab.com/alexakimenko/mlops_infra
Key features:
- MLFlow server with NGINX for authentication
- Postgres DB as MLFlow backend store
- pgAdmin server for DB administration
- MinIO server as object storage for training artifact
- Prometheus server for monitoring
- Grafana server for dashboards
- Docker for application delivery
- Gitlab CI for CI/CD