kubeflow

• Bug fixes for pipelines component

• Added Pipelines component

• Fixed TFJob Dashboard UI
• Fixed Katib manifests

1. iap setup script fix
2. prometheus config fix

Fixes:

• Pin jupyter-console to 6.0.0
• Fixed TFMA and TFDV versioning issues
• Updated Katib suggestion images
• Fixes tensorboard generated under kubeflow/tensorboard/prototypes/tensorboard-aws.jsonnet is incomplete

Initial stable release for version 0.3.0

These fixes were cherry-picked today:

1. Fix for central dashboard
2. Fix for jupyterhub UI (only 1.8 showing)
3. Added TF notebooks 1.9.0 and 1.10.1
4. TF serving images
5. Argo v2.2

Cherry-picked #1580

Updated ksonnet components for:

• CentralDashboard
• Jupyter notebooks
• TFJob operator
• Pytorch operator
• Katib
• Kubebench
• Chainer

Upgrade default GKE master version to 1.9.7-gke.6 since 1.9.7-gke.5 is no longer available.

Initial version of release v0.3.0

Fix two issues with GKE configs

1. Add read only scope to VM service account so that images can be pulled from private registries
2. Make deploy.sh work when the DM configs already exist

• Update TFJob operator image to the latest

• Upgrade GKE version to fix #1348
• Make deploy.sh point to the configs on the v0.2-branch

• Upgrade GKE version to fix #1348
• Make deploy.sh point to the configs on the v0.2-branch

Fixes

• Fix bug in the deploy scripts; #1201

Fixes in 0.2.1

• Upgrade central-ui image to v0.2.1 which includes link to JupyterHub
• Update Jupyter notebooks to include fixes to make TFMA widgets work
• Add deployment scripts for GKE and vanilla K8s
• Use PV for /home/jovyan by default
• Put all of PyTorch component into prototype

Improvements in 0.2

• TFJob - v1alpha2 - Makes a number of improvements to TFJob operator including the API
• Central UI - To make it easier to navigate among components
• JupyterHub - Better error reporting and monitoring along with support for mounting your home directory on PV for durability.
• Simplified deployment with a script to deploy in one command to an existing cluster
• GKE - One command deployment using a script using deployment manager.
  • Automatic provisioning of DNS names as .endpoints..cloud.goog
  • Simplified IAP setup
  • Support for GKE's new Stackdriver logging agents
• Simplified ksonnet prototypes for TFJob; make it easier to do advanced customization

New in 0.2.

• Katib (experimental) - hyperparameter tuning system
• PyTorch(alpha) - K8s operator for Py torch jobs
• Horvod / MPI support(alpha)

Fixes

• Upgrade central-ui image to v0.2.1 which includes link to JupyterHub
• Update Jupyter notebooks to include fixes to make TFMA widgets work
• Add deployment scripts for GKE and vanilla K8s

Stable release 0.2.0.