Kubeflow 2026 State of the Union: Orchestrating the Full ML Lifecycle
summary
Kubeflow reports concrete platform advances: a unified Kubeflow SDK unified clients (trainer, optimizer, Spark, pipelines, model registry) that supports local development and Docker-based debug runs, Trainer v2.2 with JAX and XGBoost plus Flux runtime, runtime-patch and active-deadline APIs, and planned native Katib optimization integration; Pipelines adds secret injection, cross-namespace artifact restrictions, GitOps pipeline/version APIs and a DSL.notebook component for packaging notebooks; KServe delivers serverless autoscaling and an inference-graph for multi-step LLM/generative workloads.
AI-written summary · about
chapters (9)
- 0:00QFlow and ML lifecycle intro
- 3:32Kubeflow Notebooks 2.0
- 5:58Kubeflow SDK Python
- 8:29Kubeflow Pipelines features
- 13:52Kubeflow Trainer platform
- 20:56Katib hyperparameter tuning
- 21:16KServe model inference
- 23:24Kubeflow ecosystem involvement
- 25:11Q&A MLflow integration
related talks
14mKubeflow: Where ML Meets Kubernetes
18mKubeflow Ecosystem: Past, Present, Futu...
44mEngaging the Kubeflow Community: Building an Enterprise-Ready AI/ML Platform
32mKubeflow Deep Dive
32mKubeflow Ecosystem: What’s Next for Cloud Native AI/ML and LLMOps
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
30mKubeflow, Your ML Tools on Kubernetes
08mWelcome + Opening Remarks