Hummingbird-V2 Copyright 2026 Hummingbird contributors The Hummingbird-V2 model package is licensed under the Apache License, Version 2.0. Third-party training datasets retain their own terms. The release includes attribution and provenance in TRAINING_DATA.md and training/corpus_contract.yaml. The training mix includes material from FineWeb-Edu, SmolLM-Corpus (FineWeb-Edu-Dedup and Cosmopedia v2), FineMath, DCLM baseline 1.0, Dolma 3 Dolmino, and FinePDFs-Edu. The corpus contract records the pinned source revisions and the dataset licenses declared for this training run.