Scaling Open Source ML: How Wolt Uses K8s To Deliver Great Food to Mill...
summary
Wolt runs an end-to-end MLOps platform on Kubernetes serving ML for 12M+ users, integrating Flyte for orchestration, MLflow autolog for experiment tracking, and Seldon Core inference graphs for production microservices. Dedicated k8s clusters, CI/CD templates, and a yaml-to-terraform helper provide fast, repeatable launches; deployments support shadow deployments and A/B traffic, automatic logging into Kafka/Snowflake, Prometheus metrics, and rollout automation. One production model uses GPUs today; roadmap includes hiding kubectl and open‑sourcing the yaml tool.
AI-written summary · about
chapters (20)
- 0:00Intro
- 3:19Machine Learning Life Cycle
- 4:42Challenges
- 6:39What Wolt wants
- 9:16How did we start
- 10:08Deploying on Kubernetes
- 10:38Focus
- 11:22Wolt Platform
- 12:26ML Flow
- 13:48What is ScalaCore
- 14:49Complex inference graphs
- 16:11Integrations
- 16:52Demo
- 19:14Classification
- 19:30ML Flow UI
- 27:08Scaling ML Platform
- 28:36Big ML Model
- 28:46Kafka Snowflake
- 29:22Future Work
- 29:52Questions
related talks
25mTraining and Ensuring Reliability of ML Models at Wolt: The Power of ArgoCD, Flyte, and Argo Workflows
49mKubernetes and MLOps for Scalable and Reproducible Generative AI
36mDataOps and MLOps: NLP Development and Deployment in Kubernetes made easy
26mMachine Learning Meets K8s
26mKubernetes and MLOps for Scalable and Reproducible Generative AI
34mServing ML Models at Scale with Seldon and Kubeflow
06mEvolving with Kubernetes: Embracing Model Ops
28mMachine Learning on Kubernetes at Shell: A Kubeflow Journey