{
  "family": "qwen3_tts",
  "display_name": "Qwen3-TTS",
  "description": "Qwen TTS family for controllable 10-language speech synthesis, including 3-second voice cloning, CustomVoice instruction control over preset timbres, and VoiceDesign from natural-language descriptions.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts",
    "clone",
    "design"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "zh",
    "en",
    "ja",
    "ko",
    "de",
    "fr",
    "ru",
    "pt",
    "es",
    "it"
  ],
  "capabilities": {
    "clone": [
      "speaker_reference"
    ],
    "design": [
      "voice_design"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Target synthesis language code or label; default auto lets the model infer from text and voice settings.",
        "required": false,
        "default": "auto"
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated codec frames; default comes from generation_config.json, 2048 when absent.",
        "required": false,
        "min": 1,
        "default": 2048
      },
      {
        "name": "do_sample",
        "type": "bool",
        "description": "Sample the main codec stream instead of greedy decoding; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Main codec sampling temperature; must be positive when sampling, default 0.9.",
        "required": false,
        "min": 0.0,
        "default": 0.9
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "Main codec top-k sampling limit; default 50, 0 disables top-k filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "Main codec nucleus sampling limit; default 1.0.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 1.0
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "Positive repetition penalty for the main codec stream; default 1.05.",
        "required": false,
        "min": 0.0,
        "default": 1.05
      },
      {
        "name": "codec_do_sample",
        "type": "bool",
        "description": "Sample auxiliary codec codebooks instead of greedy decoding; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "codec_temperature",
        "type": "float",
        "description": "Auxiliary codec sampling temperature; must be positive when sampling, default 0.9.",
        "required": false,
        "min": 0.0,
        "default": 0.9
      },
      {
        "name": "codec_top_k",
        "type": "int",
        "description": "Auxiliary codec top-k sampling limit; default 50, 0 disables top-k filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "codec_top_p",
        "type": "float",
        "description": "Auxiliary codec nucleus sampling limit; default 1.0.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 1.0
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Codec sampling seed; omitted requests choose a random seed.",
        "required": false,
        "min": 0
      },
      {
        "name": "reference_text",
        "type": "string",
        "description": "Transcript matching the reference audio for Base voice cloning.",
        "required": false
      },
      {
        "name": "speaker_embedding_only",
        "type": "bool",
        "description": "Use only the speaker embedding for Base voice cloning instead of ICL reference codec prompting; default false.",
        "required": false,
        "default": false
      },
      {
        "name": "instruction",
        "type": "string",
        "description": "Natural-language voice or style instruction for VoiceDesign and CustomVoice variants.",
        "required": false
      },
      {
        "name": "speaker",
        "type": "string",
        "description": "Named preset speaker for CustomVoice variants.",
        "required": false
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum UTF-8 codepoints per long-form text chunk; default 8192.",
        "required": false,
        "min": 1,
        "default": 8192
      }
    ],
    "session": [
      {
        "name": "mem_saver",
        "type": "bool",
        "description": "Release the talker cached-step graph after each request; default false.",
        "required": false,
        "default": false
      },
      {
        "name": "perf_mode",
        "type": "enum",
        "description": "Q8_0-only attention performance mode; default off keeps the exact-safe path.",
        "preset": "perf_mode_flash_attention",
        "required": false,
        "default": "off"
      },
      {
        "name": "voice_prompt_cache_slots",
        "type": "int",
        "description": "Voice prompt cache slots; default 1.",
        "required": false,
        "min": 0,
        "default": 1
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared talker matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "talker_weight_type",
        "type": "enum",
        "description": "Talker weight storage type; defaults to weight_type when set, otherwise native.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "speech_encoder_weight_type",
        "type": "enum",
        "description": "Reference speech encoder matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "speech_decoder_weight_type",
        "type": "enum",
        "description": "Speech decoder matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "conv_weight_type",
        "type": "enum",
        "description": "Speech tokenizer and speaker-encoder convolution weight storage type; default f32.",
        "preset": "weight_type_conv",
        "required": false,
        "default": "f32"
      },
      {
        "name": "talker_graph_arena_mb",
        "type": "int",
        "description": "Talker graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "speech_encoder_graph_arena_mb",
        "type": "int",
        "description": "Reference speech encoder graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "speech_decoder_graph_arena_mb",
        "type": "int",
        "description": "Speech decoder graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "speaker_encoder_graph_arena_mb",
        "type": "int",
        "description": "Speaker encoder graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "talker_constant_context_mb",
        "type": "int",
        "description": "Talker constant tensor context size in MiB; default 4096.",
        "required": false,
        "min": 1,
        "default": 4096
      },
      {
        "name": "code_predictor_constant_context_mb",
        "type": "int",
        "description": "Code predictor constant tensor context size in MiB; default 1536.",
        "required": false,
        "min": 1,
        "default": 1536
      },
      {
        "name": "speech_decoder_constant_context_mb",
        "type": "int",
        "description": "Speech decoder constant tensor context size in MiB; default 1536.",
        "required": false,
        "min": 1,
        "default": 1536
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "speech_tokenizer_config": "model:speech_tokenizer/config.json"
      },
      "tensors": {
        "model_weights": {
          "source": "weights:",
          "prefix": "model_weights"
        },
        "speech_tokenizer_weights": {
          "source": "weights:",
          "prefix": "speech_tokenizer_weights"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "speech_tokenizer_config": "model:speech_tokenizer/config.json"
      },
      "tensors": {
        "model_weights": "model:model.safetensors",
        "speech_tokenizer_weights": "model:speech_tokenizer/model.safetensors"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "qwen3_tts_1_7b_base_q8_0",
      "display_name": "Qwen3 TTS 12Hz 1.7B Base Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-Base-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-Base-GGUF/qwen3-tts-12hz-1.7b-base-q8_0_v2.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-Base-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_base_bf16",
      "display_name": "Qwen3 TTS 12Hz 1.7B Base BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-Base-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-Base-GGUF/qwen3-tts-12hz-1.7b-base-bf16.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-Base-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_base_orig",
      "display_name": "Qwen3 TTS 12Hz 1.7B Base Original-Dtype GGUF",
      "format": "gguf",
      "precision": "orig",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-Base-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-Base-GGUF/qwen3-tts-12hz-1.7b-base-orig.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-Base-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_customvoice_q8_0",
      "display_name": "Qwen3 TTS 12Hz 1.7B CustomVoice Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-CustomVoice-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-CustomVoice-GGUF/qwen3-tts-12hz-1.7b-customvoice-q8_0.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-CustomVoice-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_customvoice_bf16",
      "display_name": "Qwen3 TTS 12Hz 1.7B CustomVoice BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-CustomVoice-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-CustomVoice-GGUF/qwen3-tts-12hz-1.7b-customvoice-bf16.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-CustomVoice-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_voicedesign_q8_0",
      "display_name": "Qwen3 TTS 12Hz 1.7B VoiceDesign Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF/qwen3-tts-12hz-1.7b-voicedesign-q8_0.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_voicedesign_bf16",
      "display_name": "Qwen3 TTS 12Hz 1.7B VoiceDesign BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF",
      "files": [
        "Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF/qwen3-tts-12hz-1.7b-voicedesign-bf16.gguf"
      ],
      "strip_prefix": "Qwen3-TTS-12Hz-1.7B-VoiceDesign-GGUF"
    },
    {
      "id": "qwen3_tts_1_7b_base_safetensors",
      "display_name": "Qwen3 TTS 12Hz 1.7B Base Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "Qwen3-TTS-12Hz-1.7B-Base",
      "files": [
        "config.json",
        "generation_config.json",
        "model.safetensors",
        "speech_tokenizer/config.json",
        "speech_tokenizer/model.safetensors",
        "tokenizer_config.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
      }
    },
    {
      "id": "qwen3_tts_0_6b_base_safetensors",
      "display_name": "Qwen3 TTS 12Hz 0.6B Base Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "Qwen3-TTS-12Hz-0.6B-Base",
      "files": [
        "config.json",
        "generation_config.json",
        "model.safetensors",
        "speech_tokenizer/config.json",
        "speech_tokenizer/model.safetensors",
        "tokenizer_config.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
      }
    }
  ]
}