Lightning Talk: Streamlining MLOps Pipeline With Kubeflow on Arm64 Locally
summary
Case study of running Kubeflow on a single Arm64 box using 192 CPU cores across a three-node suitcase cluster, deploying Kubeflow Pipelines and KServe to run and serve LLM/agentic workloads without large GPUs. Uses K3s/microk8s, Docker Buildx multi‑arch manifests (amd64+arm64), Oracle A1 VMs for cross‑compile, and a patched pipeline container Dockerfile; initial exec format errors get fixed by multi‑arch images. Measures latency/SLOs and demonstrates a working on‑prem setup available at the conference booth.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:09Demo setup Arm64 hardware
- 2:37ML workloads at edge
- 4:47Kubeflow Kubernetes-native
- 6:55Local Arm64 benefits ML Ops
- 7:58Multi-arch images for Kubeflow
- 9:24Community future plans
- 10:30Local to production flow
- 11:58Q&A
related talks
22mBuilding a MLOps Platform with Kubeflow on GKE
51mMachine Learning using Kubeflow and Kubernetes
54mScaling Machine Learning Pipelines in Cloud
01h 00m#MLOps for #Java Developers A Hands On Guide with #Kubeflow and #Quarkus
29mTeam Teardown: MLOps in the Cloud for IoT
30mKubeflow, Your ML Tools on Kubernetes
20mML Ops design patterns with Kubeflow Pipelines
14mLightning Talk: Leveraging Kubeflow for Next-Generation AI Healthcare Solutions