Kubernetes and MLOps for Scalable and Reproducible Generative AI
summary
Overview of concrete MLOps choices: Kubernetes + containers for portability, GPU and CPU resource requests for training, distributed frameworks like Spark for scale, and monitoring with Prometheus. Claims include 70–80% of models never reach production and a Shell case where Kubeflow pipelines cut onboarding time — self‑service onboarding reduced from months to minutes — while lowering cost and speeding training. Tooling named: Kubeflow, Argo, Docker, PyTorch/TensorFlow/Jupyter, plus reproducibility via Kubeflow captures metadata and code versions.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 3:01Definitions and concepts
- 8:01AI/ML vs traditional engineering
- 17:08Kubernetes and MLOps
- 20:05Kubeflow overview
- 23:29Best practices and resources
- 25:25Conclusion and Q&A
related talks
49mKubernetes and MLOps for Scalable and Reproducible Generative AI
31mKubernetes and MLOps for Scalable and Reproducible Generative AI
13mMLOps: From AI Research to Production-Ready Systems
44mEngaging the Kubeflow Community: Building an Enterprise-Ready AI/ML Platform
22mBuilding a MLOps Platform with Kubeflow on GKE
41mFrom Prototype to Production Building a Scalable MLOps Platform from a Reference Arch
46mDeploying scalable Machine Learning models with Kubernetes
54mScaling Machine Learning Pipelines in Cloud