MLOps, Kubeflow, and Tekton
Talkat Continuous Delivery Summit 2019 - San Diego · on Monday, 2019-11-18· at 00:15· San Diego Convention Center - Room 1, 131 W Harbor Dr·
summary
Argues for using Tekton as compiled target for Kubeflow pipelines, claiming orchestration is the immediate problem to solve rather than full ML tooling. Describes working with Kubeflow’s Python DSL and mapping pipeline steps (data prep, training, hyperparameter tuning, distributed fairing) to Tekton tasks, notes training takes most runtime, and estimates ML workload growth at ~50% per year. Mentions active CDF SIG coordinating this integration.
AI-written summary · about
related talks
20mML Ops design patterns with Kubeflow Pipelines
29mML Engineering Kubeflow
22mBuilding a MLOps Platform with Kubeflow on GKE
14mKubeflow: Where ML Meets Kubernetes
42mAI Pipelines Workflows and ML lineage using Tekton Pipelines
05mCI/CD Pipelines and Machine Learning
14mMLOps: DevOps for AI with KubeFlow Pipelines
22mEmbrace DevOps Practices to ML Pipeline Development