You can install ML Pipeline services by running:
kubectl create -f https://storage.googleapis.com/ml-pipeline/release/0.1.3/bootstrapper.yaml

Install python SDK (python 3.5 above) by running:
pip3 install https://storage.googleapis.com/ml-pipeline/release/0.1.3/kfp.tar.gz --upgrade

Changelog since v0.1.2

SDK

• Support setting GPU limit and node selector to specify GPU type (#346)
• Support Kubernetes Volume, VolumeMount and Env APIs for Container Operator(#300)
• Add option to Container Operator to mount default GCP service account credential(#430)
• SDK/Components/Python - Removed python_op in favor of python_component (#85)
• SDK/DSL - Improved compilation of dsl.Conditional (steps->dag) (#177)
• SDK/Components - Renamed dockerContainer spec to container (#323)
• SDK/Components - Renamed container.arguments to container.args (#437)
• SDK/DSL - Added support for conditions: !=, <, <=, >=, > (#309)
• SDK/DSL - Pipeline function takes direct default values rather than dsp.PipelineParam. (#110)
• SDK/Components/Python - add support for dependencies in the component image building (#219)
• Notebook - Display highlighted logs only when there are errors. (#292)

First party components:

• Reorganized files: kubeflow (#232), dataflow (#338), “local” (#357)

UI

• View pipelines for runs created from notebooks (#447).
• Support cloning runs created from notebooks (#465).