{
  "family": "qwen3_asr",
  "display_name": "Qwen3-ASR",
  "description": "Qwen ASR model family for language identification and speech recognition across 30 languages, 22 Chinese dialects, and multiple English accents, with robustness for noisy, long-form, and singing audio.",
  "category": "asr",
  "status": "supported",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "zh",
    "en",
    "yue",
    "ar",
    "de",
    "fr",
    "es",
    "pt",
    "id",
    "it",
    "ko",
    "ru",
    "th",
    "vi",
    "ja",
    "tr",
    "hi",
    "ms",
    "nl",
    "sv",
    "da",
    "fi",
    "pl",
    "cs",
    "fil",
    "fa",
    "el",
    "hu",
    "mk",
    "ro",
    "zh dialects"
  ],
  "capabilities": {
    "asr": [
      "word_timestamps",
      "vad_chunking"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "ASR language code or label. Default auto-detects the language from speech; setting a language forces the decoder prompt.",
        "required": false,
        "default": "auto"
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated transcript tokens; default comes from generation_config.json, 512 in the shipped Qwen3-ASR configs.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "return_timestamps",
        "type": "bool",
        "description": "Return word-level timestamps by running Qwen3 ForcedAligner after ASR; default false and requires forced_aligner_path when enabled.",
        "required": false,
        "default": false
      },
      {
        "name": "audio_chunk_mode",
        "type": "enum",
        "description": "Audio chunking mode; default auto uses fixed chunks normally and VAD chunks when timestamps are requested.",
        "values": [
          "auto",
          "fixed",
          "vad",
          "none"
        ],
        "required": false,
        "default": "auto"
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Maximum audio chunk duration in seconds; default 30 for transcription and 15 when timestamps or VAD chunking are used.",
        "required": false,
        "min": 0.0
      }
    ],
    "session": [
      {
        "name": "forced_aligner_path",
        "type": "path",
        "description": "Optional Qwen3 ForcedAligner model path required for return_timestamps.",
        "required": false
      },
      {
        "name": "vad_path",
        "type": "path",
        "description": "Optional Silero VAD model path for VAD chunking; default assets/framework/models/silero_vad.",
        "required": false,
        "default": "assets/framework/models/silero_vad"
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared text decoder weight storage fallback; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "audio_encoder_weight_type",
        "type": "enum",
        "description": "Audio encoder weight storage type; default native.",
        "preset": "weight_type_conv",
        "required": false,
        "default": "native"
      },
      {
        "name": "text_decoder_weight_type",
        "type": "enum",
        "description": "Text decoder matmul weight storage type; defaults to weight_type when set, otherwise native.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "audio_encoder_graph_arena_mb",
        "type": "int",
        "description": "Audio encoder graph arena size in MiB; default 128.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "text_decoder_prefill_graph_arena_mb",
        "type": "int",
        "description": "Text decoder prefill graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "text_decoder_decode_graph_arena_mb",
        "type": "int",
        "description": "Text decoder cached-step graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "text_decoder_weight_context_mb",
        "type": "int",
        "description": "Text decoder weight context size in MiB; default 64.",
        "required": false,
        "min": 1,
        "default": 64
      }
    ],
    "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"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json",
        "processor_config": "model:processor_config.json",
        "chat_template": "model:chat_template.json",
        "chat_template_jinja": "model:chat_template.jinja",
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "tokenizer_json": "model:tokenizer.json"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "tokenizer_config": "model:tokenizer_config.json"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json",
        "processor_config": "model:processor_config.json",
        "chat_template": "model:chat_template.json",
        "chat_template_jinja": "model:chat_template.jinja",
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "tokenizer_json": "model:tokenizer.json"
      },
      "tensors": {
        "weights": "model:model.safetensors"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "qwen3_asr_1_7b_q8_0",
      "display_name": "Qwen3-ASR 1.7B Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Qwen3-ASR-1.7B-GGUF",
      "files": [
        "Qwen3-ASR-1.7B-GGUF/qwen3-asr-1.7b-q8_0.gguf"
      ],
      "strip_prefix": "Qwen3-ASR-1.7B-GGUF"
    },
    {
      "id": "qwen3_asr_1_7b_f16",
      "display_name": "Qwen3-ASR 1.7B F16 GGUF",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "Qwen3-ASR-1.7B-GGUF",
      "files": [
        "Qwen3-ASR-1.7B-GGUF/qwen3-asr-1.7b-f16.gguf"
      ],
      "strip_prefix": "Qwen3-ASR-1.7B-GGUF"
    },
    {
      "id": "qwen3_asr_0_6b_q8_0",
      "display_name": "Qwen3-ASR 0.6B Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Qwen3-ASR-0.6B-GGUF",
      "files": [
        "Qwen3-ASR-0.6B-GGUF/qwen3-asr-0.6b-q8_0.gguf"
      ],
      "strip_prefix": "Qwen3-ASR-0.6B-GGUF"
    },
    {
      "id": "qwen3_asr_0_6b_f16",
      "display_name": "Qwen3-ASR 0.6B F16 GGUF",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "Qwen3-ASR-0.6B-GGUF",
      "files": [
        "Qwen3-ASR-0.6B-GGUF/qwen3-asr-0.6b-f16.gguf"
      ],
      "strip_prefix": "Qwen3-ASR-0.6B-GGUF"
    },
    {
      "id": "qwen3_asr_1_7b_safetensors",
      "display_name": "Qwen3-ASR 1.7B HF Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "Qwen3-ASR-1.7B-hf",
      "files": [
        "config.json",
        "generation_config.json",
        "model.safetensors",
        "processor_config.json",
        "tokenizer_config.json",
        "tokenizer.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "Qwen/Qwen3-ASR-1.7B-hf"
      }
    },
    {
      "id": "qwen3_asr_0_6b_safetensors",
      "display_name": "Qwen3-ASR 0.6B Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "Qwen3-ASR-0.6B",
      "files": [
        "config.json",
        "generation_config.json",
        "model.safetensors",
        "preprocessor_config.json",
        "tokenizer_config.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "Qwen/Qwen3-ASR-0.6B"
      }
    }
  ],
  "dependencies": [
    {
      "kind": "model",
      "family": "qwen3_forced_aligner",
      "scope": "session",
      "option": "forced_aligner_path",
      "required": false,
      "required_when": [
        {
          "scope": "request",
          "option_key": "return_timestamps",
          "equals": true
        }
      ]
    },
    {
      "kind": "bundled_model",
      "family": "silero_vad",
      "path": "assets/framework/models/silero_vad",
      "scope": "session",
      "option": "vad_path",
      "required": false,
      "required_when": [
        {
          "scope": "request",
          "option_key": "audio_chunk_mode",
          "equals": "vad"
        },
        {
          "scope": "request",
          "option_key": "return_timestamps",
          "equals": true
        }
      ]
    }
  ]
}