Downloads · 30 days
15
50% of all-time downloads
Onyx-Qwen3-TTS-0.6B-4bit
wabibito/Onyx-Qwen3-TTS-0.6B-4bit
Onyx-Qwen3-TTS-0.6B-4bit is a machine learning model from wabibito. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
4-bit (group 32, affine) MLX conversion of Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice for the Onyx app's on-device OnyxTTS2 engine.
All-time downloads
30
Public
Parameters
906M
916 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors916 MB · 99%
Parameter types
How the weights are stored.
U32830M · 92%
Model card
From the Hugging Face model README
Onyx-Qwen3-TTS-0.6B-4bit
4-bit (group 32, affine) MLX conversion of Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice for the Onyx app's on-device OnyxTTS2 engine.
- Talker + code predictor: decoder-layer Linears and the text embedding 4-bit; heads, codec embeddings, text projection, and norms fp16 (mixed precision).
speech_tokenizer/: codec decoder only (fp16; EMA codebook stats f32). The encoder (voice cloning) is not shipped.tokenizer.jsoncompiled from the repo'svocab.json/merges.txt(verified identical to the official Qwen2.5 fast tokenizer) plus this repo's added tokens.
License: Apache-2.0 (inherited from the upstream weights). © Qwen team, Alibaba Cloud — thank you for the open release.
- Type
- qwen3_tts
- Created
- Sep 9, 2026
- Updated
- Sep 9, 2026