Modelpublic

HamzaBrh/qwen3-tts-0.6b-q4k

sourceHugging Facemitupdated 6mo agoView on Hugging Face

0likes

Model Card

Qwen3-TTS 0.6b — 0.6b-q4k

Q4_K hierarchical scales (llama.cpp format) + INT8 CP

Quantization Config

Parameter	Value
Backbone	q4k
Code Predictor	int8
Vocoder CPU	False
Text Embed Offload	False

Performance

Metric	Value
VRAM	2023 MB
RTF	0.144 (6.9x real-time)
Weight Storage	986 MB

Quality (Seed-TTS EN)

Metric	Value
WER (EN)	2.83%

Based On

Qwen3-TTS-12Hz-0.6B-Base

Built with MegaQwen TTS