SadikNawazAI/kitten-tts-2

KittenTTS 2 — GGUF (TQ2_1 ternary)

GGUF language-model component of KittenML/kitten-tts-2, a 1.7B speech language model with in-context voice cloning. The model reads text and writes S3 codec tokens, which a vocoder turns into 24 kHz audio.

This is an exact mirror (identical SHA-256) of cpp/model-tq2_1.gguf from the upstream repo, republished here as the GGUF that Ollama pulls for hf.co/KittenML/kitten-tts-2:latest.

Note: this repository contains only the ternary GGUF language model. A complete TTS pipeline also needs the decoder/vocoder and voice references from the upstream repository. For the full Python pipeline use pip install kittenml with KittenTTS("KittenML/kitten-tts-2").

Model details

Property	Value
Base model	KittenTTS 2 (Stellon Labs)
Architecture	qwen3
Parameters	1.73B
Quantization	TQ2_1 (1.58-bit ternary body, full-precision scales)
File size	0.96 GiB (1.03 GB)
Context length	40,960 tokens
Embedding dim	2,048
Layers	28
Vocabulary	158,241
Languages	16 named voices: Arabic, Chinese, Dutch, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Turkish

The ternary packing is lossless relative to the bf16 weights for the transformer body — each value is exactly one of {-scale, 0, +scale} within 128-wide groups.

Files

File	SHA-256
model-tq2_1.gguf	c6ad7ee1dbda3fd7a2b3dc3b8b1689f4687f761b7bd6509569b5592f543548f7

Usage

Ollama

ollama run hf.co/SadikNawazAI/kitten-tts-2

Python (full upstream pipeline)

pip install kittenml

from kittenml import KittenTTS
import soundfile as sf

m = KittenTTS("KittenML/kitten-tts-2")
audio = m.generate("Hi! How are you today?", voice="WeatherPresenterCheerful")
sf.write("output.wav", audio, m.sample_rate)

Download the standalone GGUF:

hf download SadikNawazAI/kitten-tts-2 model-tq2_1.gguf --local-dir .

License

Stellon Labs Community License — see LICENSE.md and the upstream license. Review the terms for research, non-commercial, and commercial use before deploying.