Getting started

This section is for an ML engineer who wants to inspect the current bounded pretraining path without mistaking it for a released training framework.

Path

1. Prepare a source checkout with the same dependency environment used by Training CI.

2. Compose the bounded experiment without loading data or writing artifacts.

3. Understand a bounded run, including canonical-data, NVIDIA CUDA GPU, artifact, and identity prerequisites.

The configuration quickstart is local and side-effect free. A real run reads a canonical dataset and writes attempt artifacts, so it is documented separately and is not an installation smoke test.