GitHub yet-another-ai/qts: source code, export scripts, tests, and documentation
Hugging Face dsh0416/Qwen3-TTS-12Hz-0.6B-Base-QTS: exported runtime artifacts

Recommended maintenance flow:

Change behavior in the GitHub repository first.
Export artifacts from a known Git commit.
Publish only the built model files to this Hugging Face repository, preferably from the tagged GitHub Actions release workflow in yet-another-ai/qts.
Keep this model card aligned with the GitHub docs, but do not treat this repository as a second source repository.

Included Files

Expected root layout:

text

qwen3-tts-0.6b-f16.gguf
qwen3-tts-vocoder.onnx
README.md
SHA256SUMS

Notes:

qwen3-tts-vocoder.onnx is shared across all GGUF variants in this repository.
The Rust runtime in qts expects the GGUF and vocoder files to live in the same directory by default.
Not every release must ship every quantization variant.
For the current artifact set, q8_0 is the recommended default download and f16 is the reference-quality export.

Current Quantization Support

At the moment, the qts exporter supports:

Other quantization types may appear in future releases once the export and validation pipeline is ready.

See the source repository for current usage and export documentation:

Typical local layout:

text

models/
  qwen3-tts-0.6b-f16.gguf
  qwen3-tts-vocoder.onnx

Example CLI usage:

bash

cargo run -p qwen3-tts-cli -- synthesize \
  --model-dir /path/to/models \
  --text "hello" \
  --out target/hello.wav

Current source repository snapshot:

Current artifact checksums:

qwen3-tts-0.6b-f16.gguf 33255963ba6a6f19f83830db82e0db2e5729b1d8cdbc2c2635ab2d386efe4a3c
qwen3-tts-vocoder.onnx 6d5e135a3d56cd2ede0b268ca8890155d09a53ca08586571ab8d914755963034

For future releases, it is recommended to record:

Base upstream model: