[
  {
    "metadataOutputVersion" : "3.0",
    "storagePrecision" : "Float16",
    "outputSchema" : [
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 16 × 8194)",
        "shortDescription" : "",
        "shape" : "[1, 16, 8194]",
        "name" : "logits",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 16 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 16, 1024]",
        "name" : "block_hidden",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 30720 × 1 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 30720, 1, 1024]",
        "name" : "new_key_cache",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 30720 × 1 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 30720, 1, 1024]",
        "name" : "new_value_cache",
        "type" : "MultiArray"
      }
    ],
    "modelParameters" : [

    ],
    "specificationVersion" : 9,
    "mlProgramOperationTypeHistogram" : {
      "Ios18.expandDims" : 65,
      "Ios18.softmax" : 30,
      "Ios18.cos" : 1,
      "Ios18.mul" : 482,
      "Ios18.matmul" : 121,
      "Ios18.rsqrt" : 61,
      "Ios16.reduceMean" : 61,
      "Ios18.sin" : 1,
      "Ios18.gather" : 2,
      "Ios18.add" : 272,
      "Ios16.reduceSum" : 1,
      "Ios18.reshape" : 241,
      "Ios18.linear" : 211,
      "Ios18.concat" : 64,
      "Ios18.transpose" : 151,
      "Ios18.sub" : 1,
      "Ios18.cast" : 3,
      "Ios18.pow" : 61,
      "Ios18.silu" : 30,
      "Ios18.sliceByIndex" : 181,
      "Ios18.squeeze" : 120
    },
    "computePrecision" : "Mixed (Float16, Float32, Int16, Int32, UInt16)",
    "isUpdatable" : "0",
    "stateSchema" : [

    ],
    "availability" : {
      "macOS" : "15.0",
      "tvOS" : "18.0",
      "visionOS" : "2.0",
      "watchOS" : "11.0",
      "iOS" : "18.0",
      "macCatalyst" : "18.0"
    },
    "modelType" : {
      "name" : "MLModelType_mlProgram"
    },
    "userDefinedMetadata" : {
      "com.github.apple.coremltools.source_dialect" : "TorchScript",
      "com.github.apple.coremltools.source" : "torch==2.6.0",
      "com.github.apple.coremltools.version" : "8.3.0"
    },
    "inputSchema" : [
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Int32",
        "formattedType" : "MultiArray (Int32 1 × 16)",
        "shortDescription" : "",
        "shape" : "[1, 16]",
        "name" : "speech_token_ids",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Int32",
        "formattedType" : "MultiArray (Int32 1 × 16)",
        "shortDescription" : "",
        "shape" : "[1, 16]",
        "name" : "speech_pos_ids",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 1 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 1, 1024]",
        "name" : "shift_ctx",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 30720 × 1 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 30720, 1, 1024]",
        "name" : "key_cache",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 30720 × 1 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 30720, 1, 1024]",
        "name" : "value_cache",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 1 × 1024)",
        "shortDescription" : "",
        "shape" : "[1, 1024]",
        "name" : "key_padding_mask",
        "type" : "MultiArray"
      },
      {
        "hasShapeFlexibility" : "0",
        "isOptional" : "0",
        "dataType" : "Float16",
        "formattedType" : "MultiArray (Float16 16 × 1024)",
        "shortDescription" : "",
        "shape" : "[16, 1024]",
        "name" : "block_cache_map",
        "type" : "MultiArray"
      }
    ],
    "generatedClassName" : "BlockDecoder",
    "method" : "predict"
  }
]