Modelpublic

HamzaBrh/qwen3-tts-0.6b-int8

sourceHugging Facemitupdated 6mo agoView on Hugging Face

0likes

Model Card

Qwen3-TTS 0.6b — 0.6b-int8

INT8 per-channel backbone + CP (production default)

Quantization Config

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

Performance

Metric	Value
VRAM	2082 MB
RTF	0.128 (7.8x real-time)
Weight Storage	1515 MB

Quality (Seed-TTS EN)

Metric	Value
WER (EN)	1.67%

Based On

Qwen3-TTS-12Hz-0.6B-Base

Built with MegaQwen TTS