{
  "family": "granite5asr",
  "display_name": "Granite Speech 5.0 470M TurboCTC",
  "description": "IBM Granite Speech 5.0 TurboCTC 470M parameter compact and ultra-fast English ASR model using a Conformer CTC architecture with block self-attention and self-conditioning.",
  "category": "asr",
  "status": "supported",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "en"
  ],
  "capabilities": {
    "asr": [
      "vad_chunking",
      "partial_results"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Recognition language (currently English).",
        "required": false,
        "default": "en"
      },
      {
        "name": "audio_chunk_mode",
        "type": "enum",
        "description": "Audio chunking mode: auto, fixed, or none.",
        "values": [
          "auto",
          "fixed",
          "none"
        ],
        "required": false,
        "default": "auto"
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Fixed chunk duration in seconds when not using VAD segmentation.",
        "required": false,
        "min": 0.001,
        "default": 30
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared model weight storage type.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "encoder_graph_arena_mb",
        "type": "int",
        "description": "Encoder graph arena size in MB.",
        "required": false,
        "min": 64,
        "default": 1024
      },
      {
        "name": "vad_model_path",
        "type": "string",
        "description": "Path to the Silero VAD model directory used by automatic audio chunking.",
        "required": false,
        "default": "assets/framework/models/silero_vad"
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf",
      "server",
      "stream",
      "cuda",
      "metal",
      "cpu"
    ]
  },
  "ui": {
    "recommended_package": "granite5asr_q8_0",
    "tags": [
      "ASR",
      "GGUF",
      "Stream"
    ],
    "docs": [
      "docs/asr.md",
      "docs/gguf.md"
    ],
    "summary": "IBM Granite Speech 5.0 470M TurboCTC English transcription."
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "ibm-granite/granite-speech-5.0-470m-turboctc",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "granite5asr_q8_0",
      "display_name": "Granite Speech 5.0 470M TurboCTC Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Granite-Speech-5.0-470M-TurboCTC-GGUF",
      "files": [
        "Granite-Speech-5.0-470M-TurboCTC-GGUF/granite-speech-5.0-470m-turboctc-q8_0.gguf"
      ],
      "strip_prefix": "Granite-Speech-5.0-470M-TurboCTC-GGUF",
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "audio-cpp/audio.cpp-gguf",
        "revision": "main",
        "gated": false
      }
    },
    {
      "id": "granite5asr_safetensors",
      "display_name": "Granite Speech 5.0 470M TurboCTC Native Safetensors",
      "format": "safetensors",
      "precision": "bfloat16",
      "target_directory": "granite5asr",
      "files": [
        "config.json",
        "preprocessor_config.json",
        "tokenizer.json",
        "model.safetensors"
      ]
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "tensors": {
        "weights": "model:model.safetensors"
      }
    }
  ]
}