wkwong/lunavox-qwen3-tts-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "apache-2.0",
    "base_model": "Qwen/Qwen3-TTS",
    "tags": [
      "text-to-speech",
      "tts",
      "audio",
      "cpp",
      "quantized",
      "lunavox",
      "onnx"
    ],
    "frontmatter": {
      "license": "apache-2.0",
      "base_model": "Qwen/Qwen3-TTS",
      "tags": [
        "text-to-speech",
        "tts",
        "audio",
        "cpp",
        "quantized",
        "lunavox",
        "onnx"
      ]
    },
    "hero_image_url": "",
    "summary": "This directory contains converted runtime artifacts (ONNX and GGUF) for various Qwen3-TTS model variants. These artifacts are generated from original Hugging Face checkpoints to be used by the LunaVox inference engine.",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlicense: apache-2.0\nbase_model: Qwen/Qwen3-TTS\ntags:\n- text-to-speech\n- tts\n- audio\n- cpp\n- quantized\n- lunavox\n- onnx\n---\n\n# LunaVox Runtime Models\n\nThis directory contains converted runtime artifacts (ONNX and GGUF) for various Qwen3-TTS model variants. These artifacts are generated from original Hugging Face checkpoints to be used by the LunaVox inference engine.\n\n## Downloading and Setup\n\n### 1. Automatic Source Download in `pull-model`\n`lunavox pull-model` is the only model preparation entrypoint.  \nIf required Hugging Face source weights are missing, CLI prompts in English and downloads after confirmation.\n\n### 2. Model Cache\nOriginal model weights are cached in the standard Hugging Face directory:\n`~/.cache/huggingface/hub/models--Qwen--...`\n\n## Directory Structure\n\nEach model variant subfolder (e.g., `models/base_small/`) typically contains:\n\n- `qwen3_tts_talker.q5_k.gguf`: Quantized Talker model (Llama-based).\n- `qwen3_tts_predictor.q8_0.gguf`: Quantized Predictor model (Llama-based).\n- `qwen3_tts_codec_encoder.fp16.onnx`: Audio Tokenizer (Mimi-based).\n- `qwen3_tts_speaker_encoder.fp16.onnx`: Reference Audio Speaker Encoder.\n- `qwen3_tts_decoder.fp16.onnx`: Audio Decoder (Mimi-based).\n- `embeddings/`: Projected text and codec embeddings.\n- `tokenizer.json`: Hugging Face text tokenizer configuration.\n\n## Available Variants\n- `base`: Qwen3-TTS-12Hz-1.7B-Base\n- `base_small`: Qwen3-TTS-12Hz-0.6B-Base\n- `custom`: Qwen3-TTS-12Hz-1.7B-CustomVoice\n- `custom_small`: Qwen3-TTS-12Hz-0.6B-CustomVoice\n- `design`: Qwen3-TTS-12Hz-1.7B-VoiceDesign\n",
    "related_quantizations": []
  },
  "tags": [
    "onnx",
    "gguf",
    "text-to-speech",
    "tts",
    "audio",
    "cpp",
    "quantized",
    "lunavox",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us"
  ],
  "likes": 1,
  "downloads": 607,
  "gated": false,
  "private": false,
  "last_modified": "2026-03-26T23:49:44.000Z",
  "created_at": "2025-09-24T17:52:19.000Z",
  "pipeline_tag": "text-to-speech",
  "library_name": ""
}

{
  "_id": "68d42fd3afe046d61b367f49",
  "id": "wkwong/Lunavox-Qwen3-TTS-GGUF",
  "modelId": "wkwong/Lunavox-Qwen3-TTS-GGUF",
  "sha": "694d0724541eea9e17e24d49817dc17cc08b7a34",
  "createdAt": "2025-09-24T17:52:19.000Z",
  "lastModified": "2026-03-26T23:49:44.000Z",
  "author": "wkwong",
  "downloads": 607,
  "likes": 1,
  "gated": false,
  "private": false,
  "pipeline_tag": "text-to-speech",
  "library_name": "",
  "siblings_count": 123
}