[ { "metadataOutputVersion" : "3.0", "storagePrecision" : "Mixed (Float16, Palettized (11 bits), Palettized (12 bits), Palettized (13 bits), Palettized (14 bits), Palettized (15 bits), Palettized (20 bits), Palettized (6 bits), UInt6)", "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1024)", "shortDescription" : "", "shape" : "[1, 1024]", "name" : "hidden_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "next_token", "type" : "MultiArray" } ], "modelParameters" : [ ], "specificationVersion" : 9, "functions" : [ { "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1 × 1)", "shortDescription" : "", "shape" : "[1, 1]", "name" : "token_id", "type" : "MultiArray" } ], "computePrecision" : "Mixed (Float16, Int32)", "storagePrecision" : "Mixed (Float16, Palettized (20 bits), UInt6)", "stateSchema" : [ ], "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 1024)", "shortDescription" : "", "shape" : "[1, 1, 1024]", "name" : "hidden_states", "type" : "MultiArray" } ], "name" : "embed", "mlProgramOperationTypeHistogram" : { "Ios18.greaterEqual" : 2, "Ios18.constexprLutToDense" : 1, "Ios18.add" : 2, "Select" : 2, "Ios18.gather" : 1 } }, { "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 1024)", "shortDescription" : "", "shape" : "[1, 1, 1024]", "name" : "hidden_states", "type" : "MultiArray" } ], "computePrecision" : "Mixed (Float16, Int32)", "storagePrecision" : "Mixed (Float16, Palettized (20 bits), UInt6)", "stateSchema" : [ ], "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1 × 1)", "shortDescription" : "", "shape" : "[1, 1]", "name" : "next_token", "type" : "MultiArray" } ], "name" : "lm_head_argmax", "mlProgramOperationTypeHistogram" : { "Ios18.constexprLutToDense" : 1, "Ios18.reduceArgmax" : 1, "Ios18.linear" : 1 } }, { "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1024)", "shortDescription" : "", "shape" : "[1, 1024]", "name" : "hidden", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 1 × 4 × 256)", "shortDescription" : "", "shape" : "[2, 1, 4, 256]", "name" : "q19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 1 × 256)", "shortDescription" : "", "shape" : "[2, 1, 256]", "name" : "k19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 1 × 256)", "shortDescription" : "", "shape" : "[2, 1, 256]", "name" : "v19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 2048)", "shortDescription" : "", "shape" : "[1, 1, 2048]", "name" : "mask19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "current_pos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 2048)", "shortDescription" : "", "shape" : "[1, 2048]", "name" : "gate19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 1, 1, 64]", "name" : "cos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 1, 1, 64]", "name" : "sin", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22", "type" : "MultiArray" } ], "computePrecision" : "Mixed (Float16, Float32, Int32)", "storagePrecision" : "Mixed (Float16, Palettized (11 bits), Palettized (12 bits), Palettized (13 bits), Palettized (14 bits), Palettized (15 bits), Palettized (20 bits), Palettized (6 bits), UInt6)", "stateSchema" : [ { "dataType" : "Float16", "isOptional" : "0", "formattedType" : "State (Float16 2 × 2 × 2 × 2048 × 256)", "shortDescription" : "", "shape" : "[2, 2, 2, 2048, 256]", "name" : "kv_cache", "type" : "State" } ], "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1024)", "shortDescription" : "", "shape" : "[1, 1024]", "name" : "hidden_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "next_token", "type" : "MultiArray" } ], "name" : "stateful_stage5_final_head", "mlProgramOperationTypeHistogram" : { "Ios18.expandDims" : 41, "Ios18.softmax" : 16, "Ios18.mul" : 190, "Ios18.rsqrt" : 21, "Ios16.reduceMean" : 15, "Ios18.exp" : 3, "Ios18.sigmoid" : 5, "Split" : 44, "Ios18.readState" : 5, "Ios18.add" : 101, "Ios16.reduceSum" : 12, "Ios18.writeState" : 4, "Ios18.sliceUpdate" : 4, "Ios18.reshape" : 34, "Ios18.reduceArgmax" : 1, "Ios18.constexprLutToDense" : 39, "Ios18.linear" : 36, "Ios18.conv" : 3, "Ios18.concat" : 17, "Ios16.einsum" : 80, "Ios18.transpose" : 35, "Ios18.silu" : 11, "Ios18.softplus" : 3, "Ios18.sub" : 3, "Ios18.sliceByIndex" : 28 } }, { "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 16 × 1024)", "shortDescription" : "", "shape" : "[16, 1024]", "name" : "hidden", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 16 × 4 × 256)", "shortDescription" : "", "shape" : "[2, 16, 4, 256]", "name" : "q19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 16 × 256)", "shortDescription" : "", "shape" : "[2, 16, 256]", "name" : "k19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 16 × 256)", "shortDescription" : "", "shape" : "[2, 16, 256]", "name" : "v19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 2048)", "shortDescription" : "", "shape" : "[1, 16, 2048]", "name" : "mask19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "current_pos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 16 × 2048)", "shortDescription" : "", "shape" : "[16, 2048]", "name" : "gate19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 16 × 64)", "shortDescription" : "", "shape" : "[1, 1, 16, 64]", "name" : "cos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 16 × 64)", "shortDescription" : "", "shape" : "[1, 1, 16, 64]", "name" : "sin", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22", "type" : "MultiArray" } ], "computePrecision" : "Mixed (Float16, Float32, Int32)", "storagePrecision" : "Mixed (Float16, Palettized (11 bits), Palettized (12 bits), Palettized (13 bits), Palettized (14 bits), Palettized (15 bits), Palettized (20 bits), Palettized (6 bits), UInt6)", "stateSchema" : [ { "dataType" : "Float16", "isOptional" : "0", "formattedType" : "State (Float16 2 × 2 × 2 × 2048 × 256)", "shortDescription" : "", "shape" : "[2, 2, 2, 2048, 256]", "name" : "kv_cache", "type" : "State" } ], "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 16 × 1024)", "shortDescription" : "", "shape" : "[16, 1024]", "name" : "hidden_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "next_token", "type" : "MultiArray" } ], "name" : "prefill16", "mlProgramOperationTypeHistogram" : { "Ios18.expandDims" : 308, "Ios18.softmax" : 16, "Ios18.mul" : 415, "Ios18.rsqrt" : 21, "Ios16.reduceMean" : 15, "Ios18.exp" : 48, "Ios18.sigmoid" : 5, "Split" : 44, "Ios18.readState" : 5, "Ios18.add" : 146, "Ios16.reduceSum" : 102, "Ios18.writeState" : 4, "Ios18.sliceUpdate" : 4, "Ios18.reshape" : 34, "Ios18.reduceArgmax" : 1, "Ios18.constexprLutToDense" : 39, "Ios18.linear" : 36, "Ios18.conv" : 3, "Ios18.concat" : 17, "Ios16.einsum" : 80, "Ios18.transpose" : 35, "Ios18.silu" : 11, "Ios18.softplus" : 3, "Ios18.sub" : 48, "Stack" : 3, "Ios18.sliceByIndex" : 254 } }, { "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 4 × 1024)", "shortDescription" : "", "shape" : "[4, 1024]", "name" : "hidden", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 4 × 4 × 256)", "shortDescription" : "", "shape" : "[2, 4, 4, 256]", "name" : "q19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 4 × 256)", "shortDescription" : "", "shape" : "[2, 4, 256]", "name" : "k19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 4 × 256)", "shortDescription" : "", "shape" : "[2, 4, 256]", "name" : "v19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 4 × 2048)", "shortDescription" : "", "shape" : "[1, 4, 2048]", "name" : "mask19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "current_pos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 4 × 2048)", "shortDescription" : "", "shape" : "[4, 2048]", "name" : "gate19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 4 × 64)", "shortDescription" : "", "shape" : "[1, 1, 4, 64]", "name" : "cos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 4 × 64)", "shortDescription" : "", "shape" : "[1, 1, 4, 64]", "name" : "sin", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22", "type" : "MultiArray" } ], "computePrecision" : "Mixed (Float16, Float32, Int32)", "storagePrecision" : "Mixed (Float16, Palettized (11 bits), Palettized (12 bits), Palettized (13 bits), Palettized (14 bits), Palettized (15 bits), Palettized (20 bits), Palettized (6 bits), UInt6)", "stateSchema" : [ { "dataType" : "Float16", "isOptional" : "0", "formattedType" : "State (Float16 2 × 2 × 2 × 2048 × 256)", "shortDescription" : "", "shape" : "[2, 2, 2, 2048, 256]", "name" : "kv_cache", "type" : "State" } ], "outputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 4 × 1024)", "shortDescription" : "", "shape" : "[4, 1024]", "name" : "hidden_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22_out", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "next_token", "type" : "MultiArray" } ], "name" : "prefill4", "mlProgramOperationTypeHistogram" : { "Ios18.expandDims" : 92, "Ios18.softmax" : 16, "Ios18.mul" : 235, "Ios18.rsqrt" : 21, "Ios16.reduceMean" : 15, "Ios18.exp" : 12, "Ios18.sigmoid" : 5, "Split" : 44, "Ios18.readState" : 5, "Ios18.add" : 110, "Ios16.reduceSum" : 30, "Ios18.writeState" : 4, "Ios18.sliceUpdate" : 4, "Ios18.reshape" : 34, "Ios18.reduceArgmax" : 1, "Ios18.constexprLutToDense" : 39, "Ios18.linear" : 36, "Ios18.conv" : 3, "Ios18.concat" : 17, "Ios16.einsum" : 80, "Ios18.transpose" : 35, "Ios18.silu" : 11, "Ios18.softplus" : 3, "Ios18.sub" : 12, "Stack" : 3, "Ios18.sliceByIndex" : 74 } } ], "mlProgramOperationTypeHistogram" : { "Ios18.expandDims" : 41, "Ios18.softmax" : 16, "Ios18.mul" : 190, "Ios18.rsqrt" : 21, "Ios16.reduceMean" : 15, "Ios18.exp" : 3, "Ios18.sigmoid" : 5, "Split" : 44, "Ios18.readState" : 5, "Ios18.add" : 101, "Ios16.reduceSum" : 12, "Ios18.writeState" : 4, "Ios18.sliceUpdate" : 4, "Ios18.reshape" : 34, "Ios18.reduceArgmax" : 1, "Ios18.constexprLutToDense" : 39, "Ios18.linear" : 36, "Ios18.conv" : 3, "Ios18.concat" : 17, "Ios16.einsum" : 80, "Ios18.transpose" : 35, "Ios18.silu" : 11, "Ios18.softplus" : 3, "Ios18.sub" : 3, "Ios18.sliceByIndex" : 28 }, "isUpdatable" : "0", "stateSchema" : [ { "dataType" : "Float16", "isOptional" : "0", "formattedType" : "State (Float16 2 × 2 × 2 × 2048 × 256)", "shortDescription" : "", "shape" : "[2, 2, 2, 2048, 256]", "name" : "kv_cache", "type" : "State" } ], "availability" : { "macOS" : "15.0", "tvOS" : "18.0", "visionOS" : "2.0", "watchOS" : "11.0", "iOS" : "18.0", "macCatalyst" : "18.0" }, "computePrecision" : "Mixed (Float16, Float32, Int32)", "modelType" : { "name" : "MLModelType_mlProgram" }, "inputSchema" : [ { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1024)", "shortDescription" : "", "shape" : "[1, 1024]", "name" : "hidden", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 1 × 4 × 256)", "shortDescription" : "", "shape" : "[2, 1, 4, 256]", "name" : "q19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 1 × 256)", "shortDescription" : "", "shape" : "[2, 1, 256]", "name" : "k19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 2 × 1 × 256)", "shortDescription" : "", "shape" : "[2, 1, 256]", "name" : "v19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 2048)", "shortDescription" : "", "shape" : "[1, 1, 2048]", "name" : "mask19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Int32", "formattedType" : "MultiArray (Int32 1)", "shortDescription" : "", "shape" : "[1]", "name" : "current_pos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 2048)", "shortDescription" : "", "shape" : "[1, 2048]", "name" : "gate19", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 1, 1, 64]", "name" : "cos", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 64)", "shortDescription" : "", "shape" : "[1, 1, 1, 64]", "name" : "sin", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec20", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec21", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 6144 × 3)", "shortDescription" : "", "shape" : "[1, 6144, 3]", "name" : "conv22", "type" : "MultiArray" }, { "hasShapeFlexibility" : "0", "isOptional" : "0", "dataType" : "Float16", "formattedType" : "MultiArray (Float16 1 × 16 × 128 × 128)", "shortDescription" : "", "shape" : "[1, 16, 128, 128]", "name" : "rec22", "type" : "MultiArray" } ], "defaultFunctionName" : "stateful_stage5_final_head", "generatedClassName" : "mtp_head_final_fused", "userDefinedMetadata" : { }, "method" : "predict" } ]