Automatic Speech Recognition
MLX
English
parakeet_tdt_five_value
apple-silicon
speech-to-text
asr
stt
low-bit
quantization-aware-training
on-device
Eval Results
Instructions to use FermionResearch/Phonon-2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use FermionResearch/Phonon-2 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download FermionResearch/Phonon-2 --local-dir Phonon-2
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download packed_manifest.json from FermionResearch/Phonon-2: direct link, hf CLI and curl.
- Browser
- Download file 699 Bytes
-
https://huggingface.co/FermionResearch/Phonon-2/resolve/main/packed_manifest.json
- Command line
-
hf download hf://FermionResearch/Phonon-2/packed_manifest.json
-
curl -L -o packed_manifest.json https://huggingface.co/FermionResearch/Phonon-2/resolve/main/packed_manifest.json
699 Bytes
| { | |
| "format": "fermion-five-value-parakeet-v1", | |
| "status": "PASS", | |
| "container": "model.fermion", | |
| "container_sha256": "4b6bfa3a12cc3c4e0a54f2ab3ec4ca7a842b09e5c7ecfc8e7ca0ac6cc8c11468", | |
| "container_bytes": 177438361, | |
| "five_value_modules": 264, | |
| "dense_tables": "int6", | |
| "bits_per_plane": 2, | |
| "planes": 2, | |
| "base_model": "nvidia/parakeet-tdt-0.6b-v3", | |
| "license": "cc-by-4.0", | |
| "gates": { | |
| "G-PLANE": "max abs delta 0.0 on 264/264 (mlx dequantize vs exact five-value expansion), 2026-09-25", | |
| "G-LOAD": "explicit key-set check, 697/697 names, 0 missing / 0 unexpected", | |
| "G-DECODE mlx-vs-mlx": "40/40 LibriSpeech utterances identical text and tokens (bf16 packed vs fp32 dense), 2026-09-26" | |
| } | |
| } | |