voice_audio_multi/model_specs/auk.json

{
  "schema_version": 1,
  "family": "auk",
  "display_name": "AuK",
  "description": "Tencent AuK base and AuK-Flash speech generation and editing models.",
  "category": "tts",
  "status": "experimental",
  "tasks": [
    "tts",
    "clone",
    "edit"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "auto"
  ],
  "runtime": {
    "tags": [
      "cuda",
      "gguf"
    ]
  },
  "ui": {
    "recommended_package": "auk_base_f32",
    "tags": [
      "TTS",
      "Edit"
    ],
    "docs": [
      "docs/models/auk.md"
    ]
  },
  "capabilities": {
    "tts": [
      "speaker_reference"
    ],
    "clone": [
      "speaker_reference"
    ],
    "edit": [
      "prompt_editing"
    ]
  },
  "options": {
    "request": [
      {
        "name": "instruct",
        "type": "string",
        "description": "TTS voice description. Without it, text carries the complete upstream instruction; editing instructions always go in text.",
        "required": false
      },
      {
        "name": "duration_sec",
        "type": "float",
        "description": "Positive output duration in seconds. Required without reference audio; otherwise defaults to the valid reference duration.",
        "required": false
      },
      {
        "name": "num_inference_steps",
        "type": "int",
        "description": "Euler sampling steps for AuK base; AuK-Flash always uses its fixed four-step schedule.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "guidance_scale",
        "type": "float",
        "description": "Batched guidance strength for AuK base; AuK-Flash always disables classifier-free guidance.",
        "required": false,
        "default": 2.0
      },
      {
        "name": "sway_sampling_coef",
        "type": "float",
        "description": "Sway coefficient for AuK base; ignored by AuK-Flash.",
        "required": false,
        "default": -1.0
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Nonnegative noise seed, or -1 for a random seed.",
        "required": false,
        "min": -1,
        "default": -1
      }
    ],
    "session": [
      {
        "name": "variant",
        "type": "enum",
        "values": ["base", "flash"],
        "description": "AuK generation schedule; select flash with an AuK-Flash component.",
        "required": false,
        "default": "base"
      },
      {
        "name": "model_gguf",
        "type": "string",
        "description": "AuK base or Flash generator GGUF relative to the model root; defaults to the selected variant's F32 file.",
        "required": false
      },
      {
        "name": "qwen_gguf",
        "type": "string",
        "description": "Qwen2.5-Omni-3B conditioning GGUF relative to the model root.",
        "required": false,
        "default": "qwen2.5-omni-3b-bf16.gguf"
      },
      {
        "name": "vae_gguf",
        "type": "string",
        "description": "AuK VAE GGUF relative to the model root.",
        "required": false,
        "default": "auk-vae-f32.gguf"
      },
      {
        "name": "attention",
        "type": "enum",
        "values": ["auto", "flash", "eager"],
        "description": "Flow-model attention mode.",
        "required": false,
        "default": "auto"
      },
      {
        "name": "mem_saver",
        "type": "bool",
        "description": "Release Qwen conditioning and generation weights between stages to reduce peak VRAM.",
        "required": false,
        "default": false
      }
    ],
    "load": []
  },
  "dependencies": [],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/AuK-Base-and-Flash-GGUF",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "auk_base_f32",
      "display_name": "AuK Base F32",
      "default": true,
      "format": "gguf",
      "precision": "f32",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-base-f32.gguf"
      ]
    },
    {
      "id": "auk_base_f16",
      "display_name": "AuK Base F16",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-base-f16.gguf"
      ]
    },
    {
      "id": "auk_base_q8_0",
      "display_name": "AuK Base Q8_0",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-base-q8_0.gguf"
      ]
    },
    {
      "id": "auk_flash_f32",
      "display_name": "AuK-Flash F32",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-flash-f32.gguf"
      ]
    },
    {
      "id": "auk_flash_f16",
      "display_name": "AuK-Flash F16",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-flash-f16.gguf"
      ]
    },
    {
      "id": "auk_flash_q8_0",
      "display_name": "AuK-Flash Q8_0",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-flash-q8_0.gguf"
      ]
    },
    {
      "id": "auk_qwen_bf16",
      "display_name": "AuK Qwen BF16",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "qwen2.5-omni-3b-bf16.gguf"
      ]
    },
    {
      "id": "auk_qwen_q8_0",
      "display_name": "AuK Qwen Q8_0",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "qwen2.5-omni-3b-q8_0.gguf"
      ]
    },
    {
      "id": "auk_vae_f32",
      "display_name": "AuK VAE F32",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "AuK-Base-and-Flash-GGUF",
      "files": [
        "config/auk-base.yaml",
        "config/auk-flash.yaml",
        "tokenizer/tokenizer.json",
        "tokenizer/tokenizer_config.json",
        "auk-vae-f32.gguf"
      ]
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": "."
      },
      "files": {
        "base_config": "model:config/auk-base.yaml",
        "flash_config": "model:config/auk-flash.yaml",
        "tokenizer_json": "model:tokenizer/tokenizer.json",
        "tokenizer_config": "model:tokenizer/tokenizer_config.json"
      }
    }
  ]
}