Economics and Best Practices of Running AI/ML Workloads on Kubernetes
summary
Kubernetes plus Kubeflow actually moves models from notebook to production: Kubeflow Fairing packages a notebook into a Docker image, runs the same training locally, on Kubeflow, or on Google AI Platform (demo shows local training finishing in 194 seconds, remote runs with streamed logs and Stackdriver links). A serverless engine, Nuclio, converts notebooks to real-time functions, supports GPU scheduling, Dask/RAPIDS, 400,000 events/sec single-process and shows a Samsung benchmark of ~4x speed vs plain containers; the pipeline demo runs end‑to‑end.
AI-written summary · about
chapters (10)
- 0:00Start
- 2:59Cloud native Al/ML platform
- 5:43TensorFlow training (TF Job)
- 6:25TensorFlow serving
- 9:29Day N: Deploy Model
- 13:12Code: With Kubeflow Fairing
- 22:20Kubernetes: One Platform, Complete ML Lifecycle
- 26:22Typical Data Science Pipeline
- 27:52Machine Learning Pipeline
- 29:26Serverless A Way To Simplify Data Science
related talks
51mMachine Learning using Kubeflow and Kubernetes
46mDeploying scalable Machine Learning models with Kubernetes
15mKeynote: Machine Learning on Kubernetes Made Easy With Kubeflow
31mKubernetes for Machine Learning: Managing and Scaling AI Workloads
42mKubernetes for Machine Learning: Managing and Scaling AI Workloads
54mScaling Machine Learning Pipelines in Cloud
17mDevOps for Private AI: A Scalable Platform Using Kubeflow
14mKubeflow: Where ML Meets Kubernetes