Modelpublic

dbrains/Qwen3-TTS-12Hz-1.7B-VoiceDesign-Q4_K_M-GGUF

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes127downloads

Model Card

Qwen3-TTS-12Hz-1.7B-VoiceDesign-Q4KM-GGUF

Q4KM quant of `Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign`. ~1.1 GB, vs ~2.4 GB for the existing Q8_0 GGUF.

Same tensor layout as the F16 / Q8_0 files in `khimaros/Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF`.

File

—Qwen3-TTS-12Hz-1.7B-VoiceDesign-Q4_K_M.gguf — Q4KM, 1.07 GB

Compatibility

A GGUF is a file format, not a runtime. This file targets qwen3-tts.cpp — either `khimaros` (upstream) or `dbrain` (perf fork). Mainline llama.cpp does not support the qwen3-tts architecture and will not load this file.

Usage

bash

qwen3-tts-server \
    --hf-repo    dbrain/Qwen3-TTS-12Hz-1.7B-VoiceDesign-Q4_K_M-GGUF \
    --hf-repo-v  khimaros/Qwen3-TTS-Tokenizer-12Hz-GGUF:F16

Note

Q4KM is a VRAM trade, not a speed one. On Ampere it's slower than Q80 (Q4K MMQ is poorly tuned for cc=86). Pick it when you need to fit the talker alongside other models on a small GPU.

Quality wise compared to Q8 across seeds its hard to hear a significant difference, but YMMV.

License

Apache-2.0 — derivative of `Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign`.