minimal functional tests for end to end - small data slice - fake model - dummy configs runs: - one train loop - one inference loop - evaluation
