Yooz Working Models
Collection
Task-specific models behind the Yooz apps: STT touch-up (Light/Quality), on-device ASR, text cleanup. Not general-purpose assistants. yooz.live • 9 items • Updated
Qwen3-ASR-1.7B-8bit
Superseded. Use
YoozLabs/Qwen3-ASR-1.7B-Swiftinstead. That repo carries the same weights plus a validated, reproducibletokenizer.json(regeneration script, SHA-256 manifest, parity checks against the Python reference) and a full model card. This repo is kept only so existing references keep resolving.
Early Yooz Labs redistribution of
mlx-community/Qwen3-ASR-1.7B-8bit
for Yooz Engine development.
Weights and config are byte-identical to the upstream checkpoint; the only
addition is a tokenizer.json for swift-transformers compatibility (an
earlier regeneration than the validated one shipped in
Qwen3-ASR-1.7B-Swift).
This is a special-purpose, on-device ASR checkpoint, not a general-purpose model. See the Yooz Working Models collection for the artifacts Yooz apps actually use.
License
Apache 2.0, inherited from
Qwen/Qwen3-ASR-1.7B and
mlx-community/Qwen3-ASR-1.7B-8bit.
No relicensing.
Contact
- Questions: dev@yooz.info
- Issues: github.com/yooz-labs/yooz-engine
- Downloads last month
- 36
Model size
0.8B params
Tensor type
BF16
·
U32 ·
Hardware compatibility
Log In to add your hardware
8-bit
Model tree for YoozLabs/Qwen3-ASR-1.7B-8bit
Base model
mlx-community/Qwen3-ASR-1.7B-8bit