ml-eng-lab Notebook Execution Sequence

A task notebook's control flow from parameters through training, ranking, and artifact checks.
Notebook execution sequence Sequence diagram showing papermill parameters, imports, data loading, training runs, evaluation, output persistence, and repository verification. Make targetcd task dir ParametersSMOKE_TEST ImportsPyPI nnx pin Dataset./data or bundled ModelsNNModel configs Trainmodel.train() Run listcurrent execution Rankvalidation metric Visualizeplots + tables Persist./runs ignored Scrubno host paths Verifierstructure + execution Invariant: committed notebook outputs should describe the current execution and avoid machine-specific absolute paths.

Execution

  • Papermill enters the task directory before running.
  • Parameters cells receive smoke-mode overrides.

Ranking

  • Current-run lists are preferred for top-run summaries.
  • Persistent registries are documented where intentional.

Output Hygiene

  • Verifier flags active outputs with stale repo or host-local paths.
  • Run directories remain ignored artifacts.