Kubernetes and MLOps for Scalable and Reproducible Generative AI
summary
Explains how Kubernetes plus MLOps tools actually run generative AI pipelines, showing Kubeflow components used for interactive Jupyter notebooks, Kubeflow pipelines, model training, serving and metadata capture for reproducibility; lists concrete pain points — data quality, GPU cost, model drift — and cites that over 70% of models never reach production and a Shell case where pipelines reduced months to days; focuses on architecture, resource management, autoscaling and monitoring so viewers know if the talk contains practical steps and working examples.
AI-written summary · about
chapters (11)
- 0:00Introduction to session
- 1:05Speaker introduction
- 2:29Agenda and treasure map
- 4:18Key definitions
- 7:31Why AI/ML differs from engineering
- 13:51From research to production
- 17:00Kubernetes and MLOps match
- 21:31Introducing Kubeflow
- 26:32Best practices compilation
- 28:38Ethical considerations
- 30:01Resources and closing
related talks
49mKubernetes and MLOps for Scalable and Reproducible Generative AI
26mKubernetes and MLOps for Scalable and Reproducible Generative AI
22mBuilding a MLOps Platform with Kubeflow on GKE
14mMLOps: DevOps for AI with KubeFlow Pipelines
42mKubernetes for Machine Learning: Managing and Scaling AI Workloads
54mEconomics and Best Practices of Running AI/ML Workloads on Kubernetes
17mDevOps for Private AI: A Scalable Platform Using Kubeflow
46mDeploying scalable Machine Learning models with Kubernetes