MLOps Pipelines to Deliver Tf.Lite Models to Distributed Compute Pools at the Edge
Talkat Kubernetes on Edge Day North America 2022 · on Tuesday, 2022-10-25· at 15:10· Room 250 ABC, Huntington Place: 1 Washington Blvd·
summary
End-to-end demo builds an automated MLOps pipeline that trains an object detector from TensorFlow Model Zoo, runs OCR to extract license plates, and exports three deployable artifacts: a standard TensorFlow model for datacenter, a TFLite build for mobile, and an optimized image for Raspberry Pi / NVIDIA Jetson. The pipeline uses Kubeflow pipelines with artifacts stored in MinIO, exports via TensorFlow Model Maker, and deploys serving with Seldon or as snaps/charms orchestrated by Juju to distributed edge pools.
AI-written summary · about
chapters (6)
- 0:00MLOps pipelines at the edge
- 2:29MLOps pipeline components
- 4:38Kubeflow and charms
- 7:05License plate recognition
- 8:08Kubeflow pipelines demo
- 10:13Q&A
related talks
29mTeam Teardown: MLOps in the Cloud for IoT
30mThe art of machine learning operations in TensorFlow ecosystem
46mAccelerating Machine Learning DevOps with Kubeflow
32mMLOps at Renault: A Generic Pipeline for Scalable Deployment
33mDeploying and Scaling Spark ML and Tensorflow AI Models
20mML Ops design patterns with Kubeflow Pipelines
42mDeep learning beyond the learning
39mOperating Deep Learning Pipelines Anywhere Using Kubeflow