Transcript cleanup tiny, experimental fine-tune by Arach, 2026.
Derived from Google BERT-Tiny (google/bert_uncased_L-2_H-128_A-2),
revision 30b0a37ccaaa32f332884b96992754e246e48c5f, Apache License 2.0.
Modifications: a new KEEP/DELETE token-classification head and full-model
fine-tuning on assistant-authored synthetic cleanup examples.
The tokenizer is redistributed unchanged.
Base model: https://huggingface.co/google/bert_uncased_L-2_H-128_A-2
Recipe code has a separate MIT license; Parakeet weights use CC-BY-4.0.