Qwen3-TTS VoiceDesign W8A8 ConvRot

Unofficial quantized derivative of Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign.

Quantization

• Format: W8A8 ConvRot
• Packed quantized weights remain quantized during inference
• Speech decoder remains FP16
• Includes unified model.safetensors
• Includes quantization metadata and tokenizer assets

Usage

This checkpoint is intended for the ComfyUI-Qwen3-TTS-Quant runtime.

It is not an official Qwen release.

Included files

• model.safetensors
• config.json
• generation_config.json
• tokenizer and processor files
• speech_tokenizer/
• quantization_manifest.json

License

The upstream model is released under Apache-2.0. Preserve the upstream license and attribution notices.