[ { "metadataOutputVersion" : "3.0", "storagePrecision" : "Float16", "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 2048 × 8 × 8)", "shortDescription" : "", "shape" : "[1, 2048, 8, 8]", "name" : "new_x", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 64 × 1 × 512)", "shortDescription" : "", "shape" : "[1, 64, 1, 512]", "name" : "new_k_cache_0", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 512, 1, 64]", "name" : "new_v_cache_0", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 64 × 1 × 512)", "shortDescription" : "", "shape" : "[1, 64, 1, 512]", "name" : "new_k_cache_1", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 512, 1, 64]", "name" : "new_v_cache_1", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 64 × 1 × 512)", "shortDescription" : "", "shape" : "[1, 64, 1, 512]", "name" : "new_k_cache_2", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 512, 1, 64]", "name" : "new_v_cache_2", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 64 × 1 × 512)", "shortDescription" : "", "shape" : "[1, 64, 1, 512]", "name" : "new_k_cache_3", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 512, 1, 64]", "name" : "new_v_cache_3", "type" : "MultiArray" } ], "modelParameters" : [ ], "specificationVersion" : 7, "mlProgramOperationTypeHistogram" : { "Concat" : 28, "Ios16.mul" : 172, "SliceByIndex" : 208, "Transpose" : 4, "Ios16.einsum" : 256, "Ios16.conv" : 28, "Ios16.add" : 144, "Ios16.realDiv" : 8, "Ios16.softmax" : 128, "Ios16.reduceL2Norm" : 8, "Ios16.reshape" : 28, "Ios16.silu" : 4 }, "computePrecision" : "Mixed (Float16, Int32)", "isUpdatable" : "0", "availability" : { "macOS" : "13.0", "tvOS" : "16.0", "visionOS" : "1.0", "watchOS" : "9.0", "iOS" : "16.0", "macCatalyst" : "16.0" }, "modelType" : { "name" : "MLModelType_mlProgram" }, "userDefinedMetadata" : { "com.github.apple.coremltools.source_dialect" : "TorchScript", "com.github.apple.coremltools.version" : "8.0b1", "com.github.apple.coremltools.source" : "torch==2.1.0" }, "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 2048 × 8 × 8)", "shortDescription" : "", "shape" : "[1, 2048, 8, 8]", "name" : "x", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 64 × 64)", "shortDescription" : "", "shape" : "[64, 64]", "name" : "cos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 64 × 64)", "shortDescription" : "", "shape" : "[64, 64]", "name" : "sin", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 512, 1, 64]", "name" : "mask", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 448 × 1 × 512)?", "shortDescription" : "", "shape" : "[1, 448, 1, 512]", "name" : "k_cache_0", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 448)?", "shortDescription" : "", "shape" : "[1, 512, 1, 448]", "name" : "v_cache_0", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 448 × 1 × 512)?", "shortDescription" : "", "shape" : "[1, 448, 1, 512]", "name" : "k_cache_1", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 448)?", "shortDescription" : "", "shape" : "[1, 512, 1, 448]", "name" : "v_cache_1", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 448 × 1 × 512)?", "shortDescription" : "", "shape" : "[1, 448, 1, 512]", "name" : "k_cache_2", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 448)?", "shortDescription" : "", "shape" : "[1, 512, 1, 448]", "name" : "v_cache_2", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 448 × 1 × 512)?", "shortDescription" : "", "shape" : "[1, 448, 1, 512]", "name" : "k_cache_3", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "1", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 448)?", "shortDescription" : "", "shape" : "[1, 512, 1, 448]", "name" : "v_cache_3", "type" : "MultiArray" } ], "generatedClassName" : "Llama_3_2_1B_Instruct_2024_10_10_23_56_41_chunk4", "method" : "predict" } ]