In this first phase, we'll implement the foundations to allow the benchmark to grow later. We'll start with:

• a couple minimal scripts for generating synthetic data, training lightgbm and inferencing
• some common libraries around metrics reporting
• a first benchmark pipeline in either AzureML SDK1.5 or SDK2.0 to run end-to-end and report metrics
• enabling both manual and orchestrated runs of the benchmark
• a build with a first set of unit tests
• documentation of the goals, developer guide and common routines

Milestone: an end-to-end run of the benchmark both locally (VM) and in the cloud, reporting a minimal set of wall time metrics, running on synthetic data, producing numbers we manually report in the repo as markdown.