Skip to content

Latest commit

 

History

History
68 lines (46 loc) · 1.54 KB

demos.md

File metadata and controls

68 lines (46 loc) · 1.54 KB
Raw

Conference Demo Steps

Pre-demo Setup

  • Clean up TFJobs, ACI, etc.
  • Clean up storage account
  • Scale up GPU nodes VMSS

Demos

  1. GOT Web App

    • show web app

  2. Build / Train model

    • show Python training code
    • show TFJob yaml
    • deploy training kubectl apply -f ./k8s/tfjob-training.yaml
    • show KF Dashboard
    • show logs
    • show Tensorboard
    • show Azure Storage with output, models, etc.

  3. Tensorflow Serving

    • talk about web app and need for serving program

    • show TF Serving yaml and k8s components

    • test API endpoint

      curl -X POST http://gotserving.domain.io:8501/v1/models/inception:predict -d "@./serving/daenerys-targaryen.json"

python serving/inception_client.py --server gotserving.domain.io:8500 --image ./serving/jon-snow.jpg

    • show web app again

  4. Hyperparameter Optimization

    • talk about the need for hyperparameter tuning

    • show Helm chart

    • talk about the need for virtual node kubectl get node

    • install TFJob chart

      helm install --name hyperparam ./hyperparameter/chart

    • show ACI's in portal

    • show Tensorboard

    • show Katib in KF Dashboard

  5. Kubeflow Pipelines

    • show pipeline code
    • show KF Dashboard & Pipelines UI
    • create Run
    • show where pipeline steps run as k8s pods
    • show Run output

  6. Jupyter Notebooks

    • open Kubeflow dashboard
    • browse to notebook and show GOT analysis
    • kubectl get pod -n kubeflow