This lesson classifies whether a support interaction will require a callback follow-up. The synthetic text keeps the signal explicit so the full pipeline remains CPU friendly and easy to inspect in tests.
Each example is tokenized into a fixed-length sequence and labeled as either no_callback or
callback. Tokens such as customer, callback, and followup are baked into the corpus so the
batch contract is easy to inspect in tests.
Run:
python -m tracks.nlp.lesson_43_toy_dialog_callback_prediction.train --device cpu --epochs 1Outputs are written to
outputs/nlp/lesson_43_toy_dialog_callback_prediction/<run_name>/.