The quandary of containerized machine learning
summary
BCG GAMMA presents a concrete solution: build a container (C1) that contains training code plus an empty prediction wrapper, run that container on EKS/kubernetes to produce a trained model, then run a second docker build that layers the model back into C1 to produce a hydrated C2 image. The team uses Source AI platform and Git for orchestration, deploys C2 as a horizontally scalable REST API prediction service, demos an Iris model working in production, and preserves strict version guarantees.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:49About BCG Gamma
- 1:32BCG Gamma examples
- 2:53What is machine learning
- 5:05The big problem
- 7:00Naive approaches
- 9:00Phase 1 container
- 9:46Phase 2 container
- 11:05Source AI
- 13:40Outro
related talks
24mDocker for Machine Learning!
35mAccelerate deep learning in the cloud with custom ML environments
18mBeyond Inference: Bringing ML into Production
38mAll You Need to Know to Build Your GPU Machine Learning Cloud [B]
22mProductive Machine Learning and Deep Learning Projects
54mRipped Machine Learning with Distributed Computing
42mCI/CD for Machine Learning
46mDeploying scalable Machine Learning models with Kubernetes