The GGUF files published at mudler/parakeet-cpp-gguf are converted from other
people's models, so each file family keeps the licence of its source model. The
parakeet.cpp code is MIT-licensed.
This table is the reference for scripts/publish_hf.py. The script has no default
licence: a model id that is missing from its LICENSES table is an error. The
check tests/python/check_publish_hf_licences.py (and
python3 scripts/publish_hf.py --check-licences) fails when this table and the
script differ, so change both together.
| Source checkpoint | Licence | Licence text | Notice to keep |
|---|---|---|---|
FermionResearch/Phonon-2.5 |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | Phonon-2.5, Copyright 2026 Fermion Research, derived from nvidia/parakeet-tdt-0.6b-v3 (c) NVIDIA Corporation. Training audio includes CHiME-6 (CC-BY-SA-4.0) and SPGISpeech (Kensho terms) |
FermionResearch/Phonon-2.5-English |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | Phonon-2.5 English, Copyright 2026 Fermion Research, derived from nvidia/parakeet-tdt-0.6b-v3 (c) NVIDIA Corporation. Training audio includes CHiME-6 (CC-BY-SA-4.0) and SPGISpeech (Kensho terms) |
moondream/parakeet-redux |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
moondream/parakeet-ultra |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/Nemotron-3-Diarization |
OpenMDW 1.1 | https://openmdw.ai/license/1-1/ | |
nvidia/nemotron-3.5-asr-streaming-0.6b |
OpenMDW 1.1 | https://openmdw.ai/license/1-1/ | |
nvidia/parakeet-ctc-0.6b |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-ctc-1.1b |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-rnnt-0.6b |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-rnnt-1.1b |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-tdt-0.6b-v2 |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-tdt-0.6b-v3 |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-tdt-1.1b |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-tdt_ctc-1.1b |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet-tdt_ctc-110m |
CC-BY-4.0 | https://creativecommons.org/licenses/by/4.0/ | |
nvidia/parakeet_realtime_eou_120m-v1 |
NVIDIA Open Model License | https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ | Licensed by NVIDIA Corporation under the NVIDIA Open Model License |
snakers4/silero-vad |
MIT | https://github.com/snakers4/silero-vad/blob/master/LICENSE | Copyright (c) 2020-present Silero Team. Released under the MIT license |
Notes:
- The Moondream Ultra and Redux files are converted here, not trained. Credit
Moondream and NVIDIA (
parakeet-tdt-0.6b-v3) when you use them. The Redux F16 and Q8_0 files are dequantized from the ternary weights. - The Phonon-2.5 files (multilingual and English) are converted here, not trained. The
weights are CC-BY-4.0 and the Fermion code is Apache-2.0. Credit Fermion Research and
NVIDIA (
parakeet-tdt-0.6b-v3) when you use them. The training audio of Phonon-2.5 includes CHiME-6 (CC-BY-SA-4.0) and SPGISpeech (Kensho terms); the upstream NOTICE is kept inNOTICE-phonon-2.5.txtin the Hugging Face repo. Whether the share-alike term of CHiME-6 reaches the trained weights is a legal question that this project does not settle. - The Silero VAD files are converted from the official ONNX model, not trained.
- Static card text for the files that
publish_hf.pydoes not convert (diarization, Moondream, Silero VAD, and the Nemotron 3.5 usage note) lives inscripts/hf_card/.