Holistic MLOps for better science
summary
Claims that MLOps practices help small teams and then shows an implementation using the open source vetiver framework to version share deploy and monitor models it demonstrates using pins for model boards, creating a vetiver model and a one-liner deploy to Posit Connect or a Docker image, generating model cards via Quarto, running statistical monitoring with stored metrics, and notes gaps like no CPU logging and no prescribed feedback loop for retraining so the tools work but some orchestration is left to the user
AI-written summary · about
chapters (12)
- 0:00Holistic MLOps introduction
- 1:05What is MLOps
- 2:29Vetiver for model deployment
- 5:00Beyond training models to operationalization
- 7:09Model versioning with pins
- 11:21Prototype data for deployment
- 12:44Model cards for reporting
- 16:17Deploying models as APIs
- 19:26Optimizing Docker containers
- 21:09Monitoring model performance
- 23:16Why vetiver is different
- 26:29Q&A
related talks
19mDemystifying MLOps
27mPractical MLOps for better models- 19mDemystifying MLOps
18mDemystifying MLOps with Vetiver
50mBuilding an MLOps strategy from the ground up
59mInside the fastest-growing trends in MLOps: VC & startup perspective
13mMLOps at Scale
26mMLOps for the rest of us- A poor man's guide to putting models in production