AI Workspaces for the Modern Era – Kubeflow Notebooks v2
summary
Kubeflow Notebooks v2 launches as a beta with a ground-up rewrite that adds Workspaces and WorkspaceKinds CRDs, decouples admin-defined image configs and pod configs, and ships redirects plus filter and activity rules to auto-pause idle GPUs. Demo shows a Tesla T4 GPU notebook, a TPU V5E with four cores, persistent volumes surviving image updates, and a distributed mini‑GPT run across eight hosts (32 accelerator cores) with 200k tokenized blocks; image lifecycle updates work without data loss.
AI-written summary · about
chapters (12)
- 0:00Kubeflow Notebooks v2 intro
- 1:02Seven years of evolution
- 3:53Limitations of v1
- 5:00User-friendly shift in v2
- 6:04Core features of v2
- 8:31Future roadmap for v2
- 10:38User experience in v2
- 12:26Demo: Workspaces
- 17:00Why v2 fits AI era
- 21:20Demo: Distributed AI training
- 25:57Status and call to action
- 27:19Q&A: GPU fragmentation
related talks
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
31mKubeflow - Data Science on Steroids
34mKubeflow: Multi-Tenant, Self-Serve, Accelerated Platform for Practitioners
36mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
29mKubeflow: TensorFlow on Kubernetes
54mEconomics and Best Practices of Running AI/ML Workloads on Kubernetes
15mKeynote: From Platforms to AI Factories. Has Kubernetes Solved It?
29mA Scalable Platform for Training and Inference Using Kubeflow