Upload 62 files

ANE/ANE-zh/KokoroAlbert.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3b66851b7a61e9b2f2fb17a5050a46f5d98acaec46b3822b65c3e509c15c94e
+size 243

ANE/ANE-zh/KokoroAlbert.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bae8190320b079bc995d69b19c2247c2e66faf5ebefe0cfb99bddf4efa98a86c
+size 433

ANE/ANE-zh/KokoroAlbert.mlmodelc/metadata.json

@@ -0,0 +1,90 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Mixed (Float16, Palettized (8 bits))",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "bert_dur",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.mul" : 13,
+      "Ios17.linear" : 73,
+      "Ios17.transpose" : 48,
+      "Ios17.sub" : 2,
+      "Ios16.constexprLutToDense" : 9,
+      "Range1d" : 1,
+      "Ios17.matmul" : 24,
+      "Shape" : 1,
+      "Ios17.gather" : 4,
+      "Ios17.add" : 38,
+      "Ios16.softmax" : 12,
+      "Ios17.expandDims" : 3,
+      "Ios16.gelu" : 12,
+      "Ios17.cast" : 6,
+      "Ios17.layerNorm" : 25,
+      "Ios17.reshape" : 48
+    },
+    "computePrecision" : "Mixed (Float16, Int16, Int32, UInt16)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-05-03",
+      "com.github.apple.coremltools.source" : "torch==2.11.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Int32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512",
+        "shapeRange" : "[[1, 1], [2, 512]]",
+        "formattedType" : "MultiArray (Int32 1 × 37)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37]",
+        "name" : "input_ids",
+        "shortDescription" : ""
+      },
+      {
+        "dataType" : "Int32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512",
+        "shapeRange" : "[[1, 1], [2, 512]]",
+        "formattedType" : "MultiArray (Int32 1 × 37)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37]",
+        "name" : "attention_mask",
+        "shortDescription" : ""
+      }
+    ],
+    "generatedClassName" : "KokoroAlbert",
+    "method" : "predict"
+  }
+]

ANE/ANE-zh/KokoroAlbert.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c3d8458f357f2e60380de284139af566c6c22e3719bb985524616d0a4f57e5b
+size 5718848

ANE/ANE-zh/KokoroAlbert.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7871b55a006dd65f8531edbbd48acefe4f554bad1182b852630641c3dfde2d79
+size 92156

ANE/ANE-zh/KokoroAlbert.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c3d8458f357f2e60380de284139af566c6c22e3719bb985524616d0a4f57e5b
+size 5718848

ANE/ANE-zh/KokoroAlbert.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "9FF918C7-3A11-4CA4-8998-080F81BAF843": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "EFC4F885-8922-4367-A207-C478D0734CCA": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "EFC4F885-8922-4367-A207-C478D0734CCA"
+}

ANE/ANE-zh/KokoroAlignment.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c84c8820e2e6f1b27229db6ce4ec2cc2817c546aa31d2d2f677d35be5445443
+size 243

ANE/ANE-zh/KokoroAlignment.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bede89325746a1b540c3350a473ceeaea39a5cea7a641c340e9013655865936
+size 484

ANE/ANE-zh/KokoroAlignment.mlmodelc/metadata.json

@@ -0,0 +1,107 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "en",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "asr",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.squeeze" : 1,
+      "Ios16.cumsum" : 1,
+      "Ios17.logicalAnd" : 1,
+      "Ios17.cast" : 4,
+      "Ios17.less" : 1,
+      "Ios17.matmul" : 2,
+      "Ios17.sliceByIndex" : 3,
+      "Ios17.concat" : 2,
+      "Ios17.greaterEqual" : 1,
+      "Ios17.sub" : 1,
+      "Ios17.expandDims" : 2
+    },
+    "computePrecision" : "Mixed (Float16, Int16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-05-03",
+      "com.github.apple.coremltools.source" : "torch==2.11.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Int32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512",
+        "shapeRange" : "[[1, 1], [2, 512]]",
+        "formattedType" : "MultiArray (Int32 1 × 37)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37]",
+        "name" : "pred_dur",
+        "shortDescription" : ""
+      },
+      {
+        "dataType" : "Float16",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512 × 640",
+        "shapeRange" : "[[1, 1], [2, 512], [640, 640]]",
+        "formattedType" : "MultiArray (Float16 1 × 37 × 640)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37, 640]",
+        "name" : "d",
+        "shortDescription" : ""
+      },
+      {
+        "dataType" : "Float16",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 512 × 2...512",
+        "shapeRange" : "[[1, 1], [512, 512], [2, 512]]",
+        "formattedType" : "MultiArray (Float16 1 × 512 × 37)",
+        "type" : "MultiArray",
+        "shape" : "[1, 512, 37]",
+        "name" : "t_en",
+        "shortDescription" : ""
+      }
+    ],
+    "generatedClassName" : "KokoroAlignment",
+    "method" : "predict"
+  }
+]

ANE/ANE-zh/KokoroAlignment.mlmodelc/model.mil

@@ -0,0 +1,54 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
+{
+    func main<ios17>(tensor<fp16, [1, ?, 640]> d, tensor<int32, [1, ?]> pred_dur, tensor<fp16, [1, 512, ?]> t_en) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"d", [1, 37, 640]}, {"pred_dur", [1, 37]}, {"t_en", [1, 512, 37]}}), ("RangeDims", {{"d", [[1, 1], [2, 512], [640, 640]]}, {"pred_dur", [[1, 1], [2, 512]]}, {"t_en", [[1, 1], [512, 512], [2, 512]]}})))] {
+            tensor<int32, []> var_19 = const()[name = tensor<string, []>("op_19"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> cum_dur_exclusive_0 = const()[name = tensor<string, []>("cum_dur_exclusive_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> cum_dur_reverse_0 = const()[name = tensor<string, []>("cum_dur_reverse_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> dur_to_fp16_dtype_0 = const()[name = tensor<string, []>("dur_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, ?]> pred_dur_to_fp16 = cast(dtype = dur_to_fp16_dtype_0, x = pred_dur)[name = tensor<string, []>("cast_3")];
+            tensor<fp16, [1, ?]> cum_dur_cast_fp16 = cumsum(axis = var_19, exclusive = cum_dur_exclusive_0, reverse = cum_dur_reverse_0, x = pred_dur_to_fp16)[name = tensor<string, []>("cum_dur_cast_fp16")];
+            tensor<fp16, [1, ?]> starts_cast_fp16 = sub(x = cum_dur_cast_fp16, y = pred_dur_to_fp16)[name = tensor<string, []>("starts_cast_fp16")];
+            tensor<int32, [1]> var_40_axes_0 = const()[name = tensor<string, []>("op_40_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, ?, 1]> var_40_cast_fp16 = expand_dims(axes = var_40_axes_0, x = starts_cast_fp16)[name = tensor<string, []>("op_40_cast_fp16")];
+            tensor<fp16, [1, 1, 2000]> frames_to_fp16 = const()[name = tensor<string, []>("frames_to_fp16"), val = tensor<fp16, [1, 1, 2000]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<bool, [1, ?, 2000]> var_41_cast_fp16 = greater_equal(x = frames_to_fp16, y = var_40_cast_fp16)[name = tensor<string, []>("op_41_cast_fp16")];
+            tensor<int32, [1]> var_43_axes_0 = const()[name = tensor<string, []>("op_43_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, ?, 1]> var_43_cast_fp16 = expand_dims(axes = var_43_axes_0, x = cum_dur_cast_fp16)[name = tensor<string, []>("op_43_cast_fp16")];
+            tensor<bool, [1, ?, 2000]> var_44_cast_fp16 = less(x = frames_to_fp16, y = var_43_cast_fp16)[name = tensor<string, []>("op_44_cast_fp16")];
+            tensor<bool, [1, ?, 2000]> var_45 = logical_and(x = var_41_cast_fp16, y = var_44_cast_fp16)[name = tensor<string, []>("op_45")];
+            tensor<bool, []> en_transpose_x_1 = const()[name = tensor<string, []>("en_transpose_x_1"), val = tensor<bool, []>(true)];
+            tensor<bool, []> en_transpose_y_1 = const()[name = tensor<string, []>("en_transpose_y_1"), val = tensor<bool, []>(false)];
+            tensor<string, []> alignment_to_fp16_dtype_0 = const()[name = tensor<string, []>("alignment_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, ?, 2000]> var_45_to_fp16 = cast(dtype = alignment_to_fp16_dtype_0, x = var_45)[name = tensor<string, []>("cast_2")];
+            tensor<fp16, [1, 640, 2000]> en_cast_fp16 = matmul(transpose_x = en_transpose_x_1, transpose_y = en_transpose_y_1, x = d, y = var_45_to_fp16)[name = tensor<string, []>("en_cast_fp16")];
+            tensor<bool, []> asr_transpose_x_0 = const()[name = tensor<string, []>("asr_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> asr_transpose_y_0 = const()[name = tensor<string, []>("asr_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 512, 2000]> asr_cast_fp16 = matmul(transpose_x = asr_transpose_x_0, transpose_y = asr_transpose_y_0, x = t_en, y = var_45_to_fp16)[name = tensor<string, []>("asr_cast_fp16")];
+            tensor<int32, [2]> var_65_begin_0 = const()[name = tensor<string, []>("op_65_begin_0"), val = tensor<int32, [2]>([0, -1])];
+            tensor<int32, [2]> var_65_end_0 = const()[name = tensor<string, []>("op_65_end_0"), val = tensor<int32, [2]>([1, 0])];
+            tensor<bool, [2]> var_65_end_mask_0 = const()[name = tensor<string, []>("op_65_end_mask_0"), val = tensor<bool, [2]>([true, true])];
+            tensor<fp16, [1, 1]> var_65_cast_fp16 = slice_by_index(begin = var_65_begin_0, end = var_65_end_0, end_mask = var_65_end_mask_0, x = cum_dur_cast_fp16)[name = tensor<string, []>("op_65_cast_fp16")];
+            tensor<string, []> var_70_to_int16_dtype_0 = const()[name = tensor<string, []>("op_70_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1]> var_65_cast_fp16_to_int16 = cast(dtype = var_70_to_int16_dtype_0, x = var_65_cast_fp16)[name = tensor<string, []>("cast_1")];
+            tensor<int16, []> T_a_cast_int16 = squeeze(x = var_65_cast_fp16_to_int16)[name = tensor<string, []>("T_a_cast_int16")];
+            tensor<string, []> T_a_cast_int16_to_int32_dtype_0 = const()[name = tensor<string, []>("T_a_cast_int16_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> concat_0_values0_0 = const()[name = tensor<string, []>("concat_0_values0_0"), val = tensor<int32, []>(1)];
+            tensor<int32, []> concat_0_values1_0 = const()[name = tensor<string, []>("concat_0_values1_0"), val = tensor<int32, []>(640)];
+            tensor<int32, []> concat_0_axis_0 = const()[name = tensor<string, []>("concat_0_axis_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> concat_0_interleave_0 = const()[name = tensor<string, []>("concat_0_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> T_a_cast_int16_to_int32 = cast(dtype = T_a_cast_int16_to_int32_dtype_0, x = T_a_cast_int16)[name = tensor<string, []>("cast_0")];
+            tensor<int32, [3]> concat_0 = concat(axis = concat_0_axis_0, interleave = concat_0_interleave_0, values = (concat_0_values0_0, concat_0_values1_0, T_a_cast_int16_to_int32))[name = tensor<string, []>("concat_0")];
+            tensor<int32, [3]> var_87_begin_0 = const()[name = tensor<string, []>("op_87_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<bool, [3]> var_87_end_mask_0 = const()[name = tensor<string, []>("op_87_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
+            tensor<fp16, [1, 640, ?]> en = slice_by_index(begin = var_87_begin_0, end = concat_0, end_mask = var_87_end_mask_0, x = en_cast_fp16)[name = tensor<string, []>("op_87_cast_fp16")];
+            tensor<int32, []> concat_1_values0_0 = const()[name = tensor<string, []>("concat_1_values0_0"), val = tensor<int32, []>(1)];
+            tensor<int32, []> concat_1_values1_0 = const()[name = tensor<string, []>("concat_1_values1_0"), val = tensor<int32, []>(512)];
+            tensor<int32, []> concat_1_axis_0 = const()[name = tensor<string, []>("concat_1_axis_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> concat_1_interleave_0 = const()[name = tensor<string, []>("concat_1_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [3]> concat_1 = concat(axis = concat_1_axis_0, interleave = concat_1_interleave_0, values = (concat_1_values0_0, concat_1_values1_0, T_a_cast_int16_to_int32))[name = tensor<string, []>("concat_1")];
+            tensor<int32, [3]> var_101_begin_0 = const()[name = tensor<string, []>("op_101_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<bool, [3]> var_101_end_mask_0 = const()[name = tensor<string, []>("op_101_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
+            tensor<fp16, [1, 512, ?]> asr = slice_by_index(begin = var_101_begin_0, end = concat_1, end_mask = var_101_end_mask_0, x = asr_cast_fp16)[name = tensor<string, []>("op_101_cast_fp16")];
+        } -> (en, asr);
+}

ANE/ANE-zh/KokoroAlignment.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e7d69128b59d615fc3d3cf85637a687235fc086b1eb136359adb11a61615f6b
+size 4128

ANE/ANE-zh/KokoroAlignment.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e5b90ecdad726428342766d68fecde858ea7218484141cde85af1672aa98279
+size 6971

ANE/ANE-zh/KokoroAlignment.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e7d69128b59d615fc3d3cf85637a687235fc086b1eb136359adb11a61615f6b
+size 4128

ANE/ANE-zh/KokoroAlignment.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "538DCB08-7D66-45AF-AB6B-C363411A8FC9": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "910C3F65-B5FE-43B5-80BA-3B9677FBA054": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "538DCB08-7D66-45AF-AB6B-C363411A8FC9"
+}

ANE/ANE-zh/KokoroNoise.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7536e2bb87ceabc3b06e8e7c44cd658e050f7c4fab89b45e3c758ff5c8f1d59
+size 243

ANE/ANE-zh/KokoroNoise.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7030a1d8a73eb6cda44c7ba9afabb6856c098b62bd728b5af2875900a5e5f155
+size 440

ANE/ANE-zh/KokoroNoise.mlmodelc/metadata.json

@@ -0,0 +1,111 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Mixed (Float32, Palettized (8 bits))",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "x_source_0",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "x_source_1",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios16.cumsum" : 1,
+      "Ios17.concat" : 1,
+      "UpsampleNearestNeighbor" : 1,
+      "Ios17.equal" : 1,
+      "Ios17.logicalAnd" : 5,
+      "Ios17.reshape" : 12,
+      "Ios17.instanceNorm" : 12,
+      "Ios17.transpose" : 5,
+      "Ios17.sin" : 13,
+      "Split" : 12,
+      "Ios17.expandDims" : 4,
+      "Ios16.avgPool" : 1,
+      "Ios17.add" : 50,
+      "Ios17.squeeze" : 3,
+      "Pad" : 1,
+      "Ios16.upsampleBilinear" : 1,
+      "Ios17.sqrt" : 1,
+      "Ios17.sub" : 4,
+      "Ios16.constexprLutToDense" : 26,
+      "Ios17.conv" : 16,
+      "Ios17.tanh" : 1,
+      "Ios17.linear" : 13,
+      "Ios17.pow" : 14,
+      "Ios17.cast" : 6,
+      "Ios17.less" : 3,
+      "Ios17.realDiv" : 1,
+      "Ios17.atan" : 1,
+      "Ios17.greater" : 3,
+      "Ios17.mul" : 53
+    },
+    "computePrecision" : "Mixed (Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-05-03",
+      "com.github.apple.coremltools.source" : "torch==2.11.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...4000",
+        "shapeRange" : "[[1, 1], [2, 4000]]",
+        "formattedType" : "MultiArray (Float32 1 × 266)",
+        "type" : "MultiArray",
+        "shape" : "[1, 266]",
+        "name" : "F0_curve",
+        "shortDescription" : ""
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 128]",
+        "name" : "style_timbre",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "KokoroNoise",
+    "method" : "predict"
+  }
+]

ANE/ANE-zh/KokoroNoise.mlmodelc/model.mil

@@ -0,0 +1,535 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
+{
+    func main<ios17>(tensor<fp32, [1, ?]> F0_curve, tensor<fp32, [1, 128]> style_timbre) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"F0_curve", [1, 266]}}), ("RangeDims", {{"F0_curve", [[1, 1], [2, 4000]]}})))] {
+            tensor<fp32, [1]> m_source_l_linear_bias = const()[name = tensor<string, []>("m_source_l_linear_bias"), val = tensor<fp32, [1]>([-0x1.e28358p-6])];
+            tensor<fp32, [1, 9]> m_source_l_linear_weight = const()[name = tensor<string, []>("m_source_l_linear_weight"), val = tensor<fp32, [1, 9]>([[-0x1.4dfed8p-4, -0x1.7b4864p-3, -0x1.7608cep-3, -0x1.6d4e54p-3, -0x1.946f4ap-4, 0x1.527ebcp-4, 0x1.66277ap-4, -0x1.900fdap-2, -0x1.1871f2p-1]])];
+            tensor<fp32, [11, 1, 20]> stft_conv_real_weight = const()[name = tensor<string, []>("stft_conv_real_weight"), val = tensor<fp32, [11, 1, 20]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp32, [11, 1, 20]> stft_conv_imag_weight = const()[name = tensor<string, []>("stft_conv_imag_weight"), val = tensor<fp32, [11, 1, 20]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024)))];
+            tensor<fp32, [256]> noise_convs_0_bias = const()[name = tensor<string, []>("noise_convs_0_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1984)))];
+            tensor<fp32, [256, 22, 12]> noise_convs_0_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [67584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3072))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70720))), name = tensor<string, []>("noise_convs_0_weight_palettized"), shape = tensor<uint32, [3]>([256, 22, 12])];
+            tensor<fp32, [1, 256, 1]> noise_res_0_alpha2_2 = const()[name = tensor<string, []>("noise_res_0_alpha2_2"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71808)))];
+            tensor<fp32, [1, 256, 1]> noise_res_0_alpha1_2 = const()[name = tensor<string, []>("noise_res_0_alpha1_2"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72896)))];
+            tensor<fp32, [1, 256, 1]> noise_res_0_alpha2_1 = const()[name = tensor<string, []>("noise_res_0_alpha2_1"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73984)))];
+            tensor<fp32, [1, 256, 1]> noise_res_0_alpha1_1 = const()[name = tensor<string, []>("noise_res_0_alpha1_1"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75072)))];
+            tensor<fp32, [1, 256, 1]> noise_res_0_alpha2_0 = const()[name = tensor<string, []>("noise_res_0_alpha2_0"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76160)))];
+            tensor<fp32, [1, 256, 1]> noise_res_0_alpha1_0 = const()[name = tensor<string, []>("noise_res_0_alpha1_0"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77248)))];
+            tensor<fp32, [512]> noise_res_0_adain1_0_fc_bias = const()[name = tensor<string, []>("noise_res_0_adain1_0_fc_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78336)))];
+            tensor<fp32, [512, 128]> noise_res_0_adain1_0_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80448))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146048))), name = tensor<string, []>("noise_res_0_adain1_0_fc_weight_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp32, [256]> noise_res_0_adain1_0_norm_bias = const()[name = tensor<string, []>("noise_res_0_adain1_0_norm_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147136)))];
+            tensor<fp32, [256]> noise_res_0_adain1_0_norm_weight = const()[name = tensor<string, []>("noise_res_0_adain1_0_norm_weight"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148224)))];
+            tensor<fp32, [256]> noise_res_0_convs1_0_bias = const()[name = tensor<string, []>("noise_res_0_convs1_0_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149312)))];
+            tensor<fp32, [512]> noise_res_0_adain2_0_fc_bias = const()[name = tensor<string, []>("noise_res_0_adain2_0_fc_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150400)))];
+            tensor<fp32, [512, 128]> noise_res_0_adain2_0_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152512))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218112))), name = tensor<string, []>("noise_res_0_adain2_0_fc_weight_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp32, [256]> noise_res_0_convs2_0_bias = const()[name = tensor<string, []>("noise_res_0_convs2_0_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219200)))];
+            tensor<fp32, [512]> noise_res_0_adain1_1_fc_bias = const()[name = tensor<string, []>("noise_res_0_adain1_1_fc_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220288)))];
+            tensor<fp32, [512, 128]> noise_res_0_adain1_1_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222400))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288000))), name = tensor<string, []>("noise_res_0_adain1_1_fc_weight_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp32, [256]> noise_res_0_convs1_1_bias = const()[name = tensor<string, []>("noise_res_0_convs1_1_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289088)))];
+            tensor<fp32, [512]> noise_res_0_adain2_1_fc_bias = const()[name = tensor<string, []>("noise_res_0_adain2_1_fc_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290176)))];
+            tensor<fp32, [512, 128]> noise_res_0_adain2_1_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292288))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357888))), name = tensor<string, []>("noise_res_0_adain2_1_fc_weight_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp32, [256]> noise_res_0_convs2_1_bias = const()[name = tensor<string, []>("noise_res_0_convs2_1_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358976)))];
+            tensor<fp32, [512]> noise_res_0_adain1_2_fc_bias = const()[name = tensor<string, []>("noise_res_0_adain1_2_fc_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(360064)))];
+            tensor<fp32, [512, 128]> noise_res_0_adain1_2_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362176))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(427776))), name = tensor<string, []>("noise_res_0_adain1_2_fc_weight_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp32, [256]> noise_res_0_convs1_2_bias = const()[name = tensor<string, []>("noise_res_0_convs1_2_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428864)))];
+            tensor<fp32, [512]> noise_res_0_adain2_2_fc_bias = const()[name = tensor<string, []>("noise_res_0_adain2_2_fc_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429952)))];
+            tensor<fp32, [512, 128]> noise_res_0_adain2_2_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(432064))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497664))), name = tensor<string, []>("noise_res_0_adain2_2_fc_weight_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp32, [256]> noise_res_0_convs2_2_bias = const()[name = tensor<string, []>("noise_res_0_convs2_2_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498752)))];
+            tensor<fp32, [128]> noise_convs_1_bias = const()[name = tensor<string, []>("noise_convs_1_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(499840)))];
+            tensor<fp32, [128, 22, 1]> noise_convs_1_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500416))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503296))), name = tensor<string, []>("noise_convs_1_weight_palettized"), shape = tensor<uint32, [3]>([128, 22, 1])];
+            tensor<fp32, [1, 128, 1]> noise_res_1_alpha2_2 = const()[name = tensor<string, []>("noise_res_1_alpha2_2"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(504384)))];
+            tensor<fp32, [1, 128, 1]> noise_res_1_alpha1_2 = const()[name = tensor<string, []>("noise_res_1_alpha1_2"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(504960)))];
+            tensor<fp32, [1, 128, 1]> noise_res_1_alpha2_1 = const()[name = tensor<string, []>("noise_res_1_alpha2_1"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505536)))];
+            tensor<fp32, [1, 128, 1]> noise_res_1_alpha1_1 = const()[name = tensor<string, []>("noise_res_1_alpha1_1"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(506112)))];
+            tensor<fp32, [1, 128, 1]> noise_res_1_alpha2_0 = const()[name = tensor<string, []>("noise_res_1_alpha2_0"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(506688)))];
+            tensor<fp32, [1, 128, 1]> noise_res_1_alpha1_0 = const()[name = tensor<string, []>("noise_res_1_alpha1_0"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507264)))];
+            tensor<fp32, [256]> noise_res_1_adain1_0_fc_bias = const()[name = tensor<string, []>("noise_res_1_adain1_0_fc_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507840)))];
+            tensor<fp32, [256, 128]> noise_res_1_adain1_0_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508928))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541760))), name = tensor<string, []>("noise_res_1_adain1_0_fc_weight_palettized"), shape = tensor<uint32, [2]>([256, 128])];
+            tensor<fp32, [128]> noise_res_1_adain1_0_norm_bias = const()[name = tensor<string, []>("noise_res_1_adain1_0_norm_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(542848)))];
+            tensor<fp32, [128]> noise_res_1_adain1_0_norm_weight = const()[name = tensor<string, []>("noise_res_1_adain1_0_norm_weight"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543424)))];
+            tensor<fp32, [128]> noise_res_1_convs1_0_bias = const()[name = tensor<string, []>("noise_res_1_convs1_0_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(544000)))];
+            tensor<fp32, [256]> noise_res_1_adain2_0_fc_bias = const()[name = tensor<string, []>("noise_res_1_adain2_0_fc_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(544576)))];
+            tensor<fp32, [256, 128]> noise_res_1_adain2_0_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545664))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578496))), name = tensor<string, []>("noise_res_1_adain2_0_fc_weight_palettized"), shape = tensor<uint32, [2]>([256, 128])];
+            tensor<fp32, [128]> noise_res_1_convs2_0_bias = const()[name = tensor<string, []>("noise_res_1_convs2_0_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579584)))];
+            tensor<fp32, [256]> noise_res_1_adain1_1_fc_bias = const()[name = tensor<string, []>("noise_res_1_adain1_1_fc_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(580160)))];
+            tensor<fp32, [256, 128]> noise_res_1_adain1_1_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(581248))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614080))), name = tensor<string, []>("noise_res_1_adain1_1_fc_weight_palettized"), shape = tensor<uint32, [2]>([256, 128])];
+            tensor<fp32, [128]> noise_res_1_convs1_1_bias = const()[name = tensor<string, []>("noise_res_1_convs1_1_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615168)))];
+            tensor<fp32, [256]> noise_res_1_adain2_1_fc_bias = const()[name = tensor<string, []>("noise_res_1_adain2_1_fc_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615744)))];
+            tensor<fp32, [256, 128]> noise_res_1_adain2_1_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(616832))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(649664))), name = tensor<string, []>("noise_res_1_adain2_1_fc_weight_palettized"), shape = tensor<uint32, [2]>([256, 128])];
+            tensor<fp32, [128]> noise_res_1_convs2_1_bias = const()[name = tensor<string, []>("noise_res_1_convs2_1_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650752)))];
+            tensor<fp32, [256]> noise_res_1_adain1_2_fc_bias = const()[name = tensor<string, []>("noise_res_1_adain1_2_fc_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651328)))];
+            tensor<fp32, [256, 128]> noise_res_1_adain1_2_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652416))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(685248))), name = tensor<string, []>("noise_res_1_adain1_2_fc_weight_palettized"), shape = tensor<uint32, [2]>([256, 128])];
+            tensor<fp32, [128]> noise_res_1_convs1_2_bias = const()[name = tensor<string, []>("noise_res_1_convs1_2_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686336)))];
+            tensor<fp32, [256]> noise_res_1_adain2_2_fc_bias = const()[name = tensor<string, []>("noise_res_1_adain2_2_fc_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686912)))];
+            tensor<fp32, [256, 128]> noise_res_1_adain2_2_fc_weight_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(688000))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(720832))), name = tensor<string, []>("noise_res_1_adain2_2_fc_weight_palettized"), shape = tensor<uint32, [2]>([256, 128])];
+            tensor<fp32, [128]> noise_res_1_convs2_2_bias = const()[name = tensor<string, []>("noise_res_1_convs2_2_bias"), val = tensor<fp32, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721920)))];
+            tensor<int32, [1]> input_1_axes_0 = const()[name = tensor<string, []>("input_1_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, ?]> input_1 = expand_dims(axes = input_1_axes_0, x = F0_curve)[name = tensor<string, []>("input_1")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp32, [1, 1, ?, 1]> expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = input_1)[name = tensor<string, []>("expand_dims_0")];
+            tensor<int32, []> upsample_nearest_neighbor_0_scale_factor_height_0 = const()[name = tensor<string, []>("upsample_nearest_neighbor_0_scale_factor_height_0"), val = tensor<int32, []>(300)];
+            tensor<int32, []> upsample_nearest_neighbor_0_scale_factor_width_0 = const()[name = tensor<string, []>("upsample_nearest_neighbor_0_scale_factor_width_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 1, ?, 1]> upsample_nearest_neighbor_0 = upsample_nearest_neighbor(scale_factor_height = upsample_nearest_neighbor_0_scale_factor_height_0, scale_factor_width = upsample_nearest_neighbor_0_scale_factor_width_0, x = expand_dims_0)[name = tensor<string, []>("upsample_nearest_neighbor_0")];
+            tensor<int32, [1]> var_26_axes_0 = const()[name = tensor<string, []>("op_26_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp32, [1, 1, ?]> var_26 = squeeze(axes = var_26_axes_0, x = upsample_nearest_neighbor_0)[name = tensor<string, []>("op_26")];
+            tensor<int32, []> var_30 = const()[name = tensor<string, []>("op_30"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 9, 1]> const_26 = const()[name = tensor<string, []>("const_26"), val = tensor<fp32, [1, 9, 1]>([[[0x1p+0], [0x1p+1], [0x1.8p+1], [0x1p+2], [0x1.4p+2], [0x1.8p+2], [0x1.cp+2], [0x1p+3], [0x1.2p+3]]])];
+            tensor<fp32, [1, 9, ?]> fn = mul(x = var_26, y = const_26)[name = tensor<string, []>("fn")];
+            tensor<fp32, []> _inversed_rad_values_y_0 = const()[name = tensor<string, []>("_inversed_rad_values_y_0"), val = tensor<fp32, []>(0x1.5d867cp-15)];
+            tensor<fp32, [1, 9, ?]> _inversed_rad_values = mul(x = fn, y = _inversed_rad_values_y_0)[name = tensor<string, []>("_inversed_rad_values")];
+            tensor<int32, [1]> var_50 = const()[name = tensor<string, []>("op_50"), val = tensor<int32, [1]>([300])];
+            tensor<int32, [1]> var_51 = const()[name = tensor<string, []>("op_51"), val = tensor<int32, [1]>([300])];
+            tensor<string, []> rv_down_pad_type_0 = const()[name = tensor<string, []>("rv_down_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> rv_down_pad_0 = const()[name = tensor<string, []>("rv_down_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<bool, []> rv_down_exclude_padding_from_average_0 = const()[name = tensor<string, []>("rv_down_exclude_padding_from_average_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> rv_down_ceil_mode_0 = const()[name = tensor<string, []>("rv_down_ceil_mode_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 9, ?]> rv_down = avg_pool(ceil_mode = rv_down_ceil_mode_0, exclude_padding_from_average = rv_down_exclude_padding_from_average_0, kernel_sizes = var_50, pad = rv_down_pad_0, pad_type = rv_down_pad_type_0, strides = var_51, x = _inversed_rad_values)[name = tensor<string, []>("rv_down")];
+            tensor<int32, [3]> rad_values_down_perm_0 = const()[name = tensor<string, []>("rad_values_down_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<bool, []> var_55_exclusive_0 = const()[name = tensor<string, []>("op_55_exclusive_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_55_reverse_0 = const()[name = tensor<string, []>("op_55_reverse_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, ?, 9]> rad_values_down = transpose(perm = rad_values_down_perm_0, x = rv_down)[name = tensor<string, []>("transpose_4")];
+            tensor<fp32, [1, ?, 9]> var_55 = cumsum(axis = var_30, exclusive = var_55_exclusive_0, reverse = var_55_reverse_0, x = rad_values_down)[name = tensor<string, []>("op_55")];
+            tensor<fp32, []> var_56 = const()[name = tensor<string, []>("op_56"), val = tensor<fp32, []>(0x1.921fb6p+2)];
+            tensor<fp32, [1, ?, 9]> phase_1 = mul(x = var_55, y = var_56)[name = tensor<string, []>("phase_1")];
+            tensor<int32, [3]> var_58_perm_0 = const()[name = tensor<string, []>("op_58_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp32, []> var_59_promoted = const()[name = tensor<string, []>("op_59_promoted"), val = tensor<fp32, []>(0x1.2cp+8)];
+            tensor<fp32, [1, 9, ?]> var_58 = transpose(perm = var_58_perm_0, x = phase_1)[name = tensor<string, []>("transpose_3")];
+            tensor<fp32, [1, 9, ?]> input_3 = mul(x = var_58, y = var_59_promoted)[name = tensor<string, []>("input_3")];
+            tensor<int32, [1]> expand_dims_1_axes_0 = const()[name = tensor<string, []>("expand_dims_1_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp32, [1, 9, ?, 1]> expand_dims_1 = expand_dims(axes = expand_dims_1_axes_0, x = input_3)[name = tensor<string, []>("expand_dims_1")];
+            tensor<int32, []> upsample_bilinear_0_scale_factor_height_0 = const()[name = tensor<string, []>("upsample_bilinear_0_scale_factor_height_0"), val = tensor<int32, []>(300)];
+            tensor<bool, []> upsample_bilinear_0_align_corners_0 = const()[name = tensor<string, []>("upsample_bilinear_0_align_corners_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> upsample_bilinear_0_scale_factor_width_0 = const()[name = tensor<string, []>("upsample_bilinear_0_scale_factor_width_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 9, ?, 1]> upsample_bilinear_0 = upsample_bilinear(align_corners = upsample_bilinear_0_align_corners_0, scale_factor_height = upsample_bilinear_0_scale_factor_height_0, scale_factor_width = upsample_bilinear_0_scale_factor_width_0, x = expand_dims_1)[name = tensor<string, []>("upsample_bilinear_0")];
+            tensor<int32, [1]> ph_up_axes_0 = const()[name = tensor<string, []>("ph_up_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp32, [1, 9, ?]> ph_up = squeeze(axes = ph_up_axes_0, x = upsample_bilinear_0)[name = tensor<string, []>("ph_up")];
+            tensor<int32, [3]> phase_perm_0 = const()[name = tensor<string, []>("phase_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp32, [1, ?, 9]> phase = transpose(perm = phase_perm_0, x = ph_up)[name = tensor<string, []>("transpose_2")];
+            tensor<fp32, [1, ?, 9]> var_64 = sin(x = phase)[name = tensor<string, []>("op_64")];
+            tensor<fp32, []> var_65 = const()[name = tensor<string, []>("op_65"), val = tensor<fp32, []>(0x1.99999ap-4)];
+            tensor<fp32, [1, ?, 9]> sines = mul(x = var_64, y = var_65)[name = tensor<string, []>("sines")];
+            tensor<fp32, []> var_31_promoted = const()[name = tensor<string, []>("op_31_promoted"), val = tensor<fp32, []>(0x1.4p+3)];
+            tensor<int32, [3]> transpose_0_perm_0 = const()[name = tensor<string, []>("transpose_0_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp32, [1, ?, 1]> transpose_0 = transpose(perm = transpose_0_perm_0, x = var_26)[name = tensor<string, []>("transpose_1")];
+            tensor<bool, [1, ?, 1]> var_67 = greater(x = transpose_0, y = var_31_promoted)[name = tensor<string, []>("op_67")];
+            tensor<string, []> uv_dtype_0 = const()[name = tensor<string, []>("uv_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, []> var_69 = const()[name = tensor<string, []>("op_69"), val = tensor<fp32, []>(0x1.89374cp-9)];
+            tensor<fp32, [1, ?, 1]> uv = cast(dtype = uv_dtype_0, x = var_67)[name = tensor<string, []>("cast_5")];
+            tensor<fp32, [1, ?, 1]> var_70 = mul(x = uv, y = var_69)[name = tensor<string, []>("op_70")];
+            tensor<fp32, []> var_30_promoted = const()[name = tensor<string, []>("op_30_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, ?, 1]> var_71 = sub(x = var_30_promoted, y = uv)[name = tensor<string, []>("op_71")];
+            tensor<fp32, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<fp32, []>(0x1.99999ap-4)];
+            tensor<fp32, [1, ?, 1]> var_73 = mul(x = var_71, y = var_72)[name = tensor<string, []>("op_73")];
+            tensor<fp32, []> _inversed_75_y_0 = const()[name = tensor<string, []>("_inversed_75_y_0"), val = tensor<fp32, []>(0x1.555556p-2)];
+            tensor<fp32, [1, ?, 1]> _inversed_75 = mul(x = var_73, y = _inversed_75_y_0)[name = tensor<string, []>("_inversed_75")];
+            tensor<fp32, [1, ?, 1]> noise_amp = add(x = var_70, y = _inversed_75)[name = tensor<string, []>("noise_amp")];
+            tensor<fp32, []> var_77 = const()[name = tensor<string, []>("op_77"), val = tensor<fp32, []>(0x1.47ae14p-7)];
+            tensor<fp32, [1, ?, 1]> noise = mul(x = noise_amp, y = var_77)[name = tensor<string, []>("noise")];
+            tensor<fp32, [1, ?, 9]> var_79 = mul(x = sines, y = uv)[name = tensor<string, []>("op_79")];
+            tensor<fp32, [1, ?, 9]> input_5 = add(x = var_79, y = noise)[name = tensor<string, []>("input_5")];
+            tensor<fp32, [1, ?, 1]> input_7 = linear(bias = m_source_l_linear_bias, weight = m_source_l_linear_weight, x = input_5)[name = tensor<string, []>("linear_0")];
+            tensor<fp32, [1, ?, 1]> har_source = tanh(x = input_7)[name = tensor<string, []>("har_source")];
+            tensor<int32, [3]> var_90_perm_0 = const()[name = tensor<string, []>("op_90_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> input_9_axes_0 = const()[name = tensor<string, []>("input_9_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, ?]> var_90 = transpose(perm = var_90_perm_0, x = har_source)[name = tensor<string, []>("transpose_0")];
+            tensor<fp32, [1, ?]> input_9 = squeeze(axes = input_9_axes_0, x = var_90)[name = tensor<string, []>("input_9")];
+            tensor<fp32, []> const_1 = const()[name = tensor<string, []>("const_1"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<int32, [4]> waveform_pad_0 = const()[name = tensor<string, []>("waveform_pad_0"), val = tensor<int32, [4]>([0, 0, 10, 10])];
+            tensor<string, []> waveform_mode_0 = const()[name = tensor<string, []>("waveform_mode_0"), val = tensor<string, []>("replicate")];
+            tensor<fp32, [1, ?]> waveform = pad(constant_val = const_1, mode = waveform_mode_0, pad = waveform_pad_0, x = input_9)[name = tensor<string, []>("waveform")];
+            tensor<int32, [1]> input_11_axes_0 = const()[name = tensor<string, []>("input_11_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, ?]> input_11 = expand_dims(axes = input_11_axes_0, x = waveform)[name = tensor<string, []>("input_11")];
+            tensor<string, []> real_out_pad_type_0 = const()[name = tensor<string, []>("real_out_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> real_out_strides_0 = const()[name = tensor<string, []>("real_out_strides_0"), val = tensor<int32, [1]>([5])];
+            tensor<int32, [2]> real_out_pad_0 = const()[name = tensor<string, []>("real_out_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> real_out_dilations_0 = const()[name = tensor<string, []>("real_out_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> real_out_groups_0 = const()[name = tensor<string, []>("real_out_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 11, ?]> real_out = conv(dilations = real_out_dilations_0, groups = real_out_groups_0, pad = real_out_pad_0, pad_type = real_out_pad_type_0, strides = real_out_strides_0, weight = stft_conv_real_weight, x = input_11)[name = tensor<string, []>("real_out")];
+            tensor<string, []> imag_out_pad_type_0 = const()[name = tensor<string, []>("imag_out_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> imag_out_strides_0 = const()[name = tensor<string, []>("imag_out_strides_0"), val = tensor<int32, [1]>([5])];
+            tensor<int32, [2]> imag_out_pad_0 = const()[name = tensor<string, []>("imag_out_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> imag_out_dilations_0 = const()[name = tensor<string, []>("imag_out_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> imag_out_groups_0 = const()[name = tensor<string, []>("imag_out_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 11, ?]> imag_out = conv(dilations = imag_out_dilations_0, groups = imag_out_groups_0, pad = imag_out_pad_0, pad_type = imag_out_pad_type_0, strides = imag_out_strides_0, weight = stft_conv_imag_weight, x = input_11)[name = tensor<string, []>("imag_out")];
+            tensor<fp32, []> var_125_promoted = const()[name = tensor<string, []>("op_125_promoted"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 11, ?]> var_126 = pow(x = real_out, y = var_125_promoted)[name = tensor<string, []>("op_126")];
+            tensor<fp32, []> var_127_promoted = const()[name = tensor<string, []>("op_127_promoted"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 11, ?]> var_128 = pow(x = imag_out, y = var_127_promoted)[name = tensor<string, []>("op_128")];
+            tensor<fp32, [1, 11, ?]> var_130 = add(x = var_126, y = var_128)[name = tensor<string, []>("op_130")];
+            tensor<fp32, []> var_132 = const()[name = tensor<string, []>("op_132"), val = tensor<fp32, []>(0x1.6849b8p-47)];
+            tensor<fp32, [1, 11, ?]> var_133 = add(x = var_130, y = var_132)[name = tensor<string, []>("op_133")];
+            tensor<fp32, [1, 11, ?]> har_spec = sqrt(x = var_133)[name = tensor<string, []>("har_spec")];
+            tensor<fp32, []> less_0_y_0 = const()[name = tensor<string, []>("less_0_y_0"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<bool, [1, 11, ?]> less_0 = less(x = imag_out, y = less_0_y_0)[name = tensor<string, []>("less_0")];
+            tensor<fp32, []> greater_0_y_0 = const()[name = tensor<string, []>("greater_0_y_0"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<bool, [1, 11, ?]> greater_0 = greater(x = imag_out, y = greater_0_y_0)[name = tensor<string, []>("greater_0")];
+            tensor<fp32, []> less_1_y_0 = const()[name = tensor<string, []>("less_1_y_0"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<bool, [1, 11, ?]> less_1 = less(x = real_out, y = less_1_y_0)[name = tensor<string, []>("less_1")];
+            tensor<fp32, []> equal_0_y_0 = const()[name = tensor<string, []>("equal_0_y_0"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<bool, [1, 11, ?]> equal_0 = equal(x = real_out, y = equal_0_y_0)[name = tensor<string, []>("equal_0")];
+            tensor<bool, [1, 11, ?]> logical_and_0 = logical_and(x = greater_0, y = less_1)[name = tensor<string, []>("logical_and_0")];
+            tensor<bool, [1, 11, ?]> logical_and_1 = logical_and(x = less_0, y = less_1)[name = tensor<string, []>("logical_and_1")];
+            tensor<bool, [1, 11, ?]> logical_and_2 = logical_and(x = greater_0, y = equal_0)[name = tensor<string, []>("logical_and_2")];
+            tensor<bool, [1, 11, ?]> logical_and_3 = logical_and(x = less_0, y = equal_0)[name = tensor<string, []>("logical_and_3")];
+            tensor<string, []> cast_5_dtype_0 = const()[name = tensor<string, []>("cast_5_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<string, []> cast_6_dtype_0 = const()[name = tensor<string, []>("cast_6_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<string, []> cast_7_dtype_0 = const()[name = tensor<string, []>("cast_7_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<string, []> cast_8_dtype_0 = const()[name = tensor<string, []>("cast_8_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, []> mul_0_y_0 = const()[name = tensor<string, []>("mul_0_y_0"), val = tensor<fp32, []>(0x1.921fb6p+1)];
+            tensor<fp32, [1, 11, ?]> cast_5 = cast(dtype = cast_5_dtype_0, x = logical_and_0)[name = tensor<string, []>("cast_4")];
+            tensor<fp32, [1, 11, ?]> mul_0 = mul(x = cast_5, y = mul_0_y_0)[name = tensor<string, []>("mul_0")];
+            tensor<fp32, []> mul_1_y_0 = const()[name = tensor<string, []>("mul_1_y_0"), val = tensor<fp32, []>(0x1.921fb6p+1)];
+            tensor<fp32, [1, 11, ?]> cast_6 = cast(dtype = cast_6_dtype_0, x = logical_and_1)[name = tensor<string, []>("cast_3")];
+            tensor<fp32, [1, 11, ?]> mul_1 = mul(x = cast_6, y = mul_1_y_0)[name = tensor<string, []>("mul_1")];
+            tensor<fp32, []> sub_0_x_0 = const()[name = tensor<string, []>("sub_0_x_0"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 11, ?]> cast_7 = cast(dtype = cast_7_dtype_0, x = logical_and_2)[name = tensor<string, []>("cast_2")];
+            tensor<fp32, [1, 11, ?]> sub_0 = sub(x = sub_0_x_0, y = cast_7)[name = tensor<string, []>("sub_0")];
+            tensor<fp32, []> mul_2_y_0 = const()[name = tensor<string, []>("mul_2_y_0"), val = tensor<fp32, []>(0x1.921fb6p+0)];
+            tensor<fp32, [1, 11, ?]> mul_2 = mul(x = cast_7, y = mul_2_y_0)[name = tensor<string, []>("mul_2")];
+            tensor<fp32, []> sub_1_x_0 = const()[name = tensor<string, []>("sub_1_x_0"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 11, ?]> cast_8 = cast(dtype = cast_8_dtype_0, x = logical_and_3)[name = tensor<string, []>("cast_1")];
+            tensor<fp32, [1, 11, ?]> sub_1 = sub(x = sub_1_x_0, y = cast_8)[name = tensor<string, []>("sub_1")];
+            tensor<fp32, []> mul_3_y_0 = const()[name = tensor<string, []>("mul_3_y_0"), val = tensor<fp32, []>(-0x1.921fb6p+0)];
+            tensor<fp32, [1, 11, ?]> mul_3 = mul(x = cast_8, y = mul_3_y_0)[name = tensor<string, []>("mul_3")];
+            tensor<fp32, []> greater_1_y_0 = const()[name = tensor<string, []>("greater_1_y_0"), val = tensor<fp32, []>(-0x1.5798eep-27)];
+            tensor<bool, [1, 11, ?]> greater_1 = greater(x = real_out, y = greater_1_y_0)[name = tensor<string, []>("greater_1")];
+            tensor<fp32, []> less_2_y_0 = const()[name = tensor<string, []>("less_2_y_0"), val = tensor<fp32, []>(0x1.5798eep-27)];
+            tensor<bool, [1, 11, ?]> less_2 = less(x = real_out, y = less_2_y_0)[name = tensor<string, []>("less_2")];
+            tensor<bool, [1, 11, ?]> logical_and_4 = logical_and(x = greater_1, y = less_2)[name = tensor<string, []>("logical_and_4")];
+            tensor<string, []> cast_9_dtype_0 = const()[name = tensor<string, []>("cast_9_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, []> mul_4_y_0 = const()[name = tensor<string, []>("mul_4_y_0"), val = tensor<fp32, []>(0x1.5798eep-26)];
+            tensor<fp32, [1, 11, ?]> cast_9 = cast(dtype = cast_9_dtype_0, x = logical_and_4)[name = tensor<string, []>("cast_0")];
+            tensor<fp32, [1, 11, ?]> mul_4 = mul(x = cast_9, y = mul_4_y_0)[name = tensor<string, []>("mul_4")];
+            tensor<fp32, [1, 11, ?]> add_0 = add(x = real_out, y = mul_4)[name = tensor<string, []>("add_0")];
+            tensor<fp32, [1, 11, ?]> real_div_0 = real_div(x = imag_out, y = add_0)[name = tensor<string, []>("real_div_0")];
+            tensor<fp32, [1, 11, ?]> atan_0 = atan(x = real_div_0)[name = tensor<string, []>("atan_0")];
+            tensor<fp32, [1, 11, ?]> add_1 = add(x = atan_0, y = mul_0)[name = tensor<string, []>("add_1")];
+            tensor<fp32, [1, 11, ?]> sub_2 = sub(x = add_1, y = mul_1)[name = tensor<string, []>("sub_2")];
+            tensor<fp32, [1, 11, ?]> mul_5 = mul(x = sub_2, y = sub_0)[name = tensor<string, []>("mul_5")];
+            tensor<fp32, [1, 11, ?]> add_2 = add(x = mul_5, y = mul_2)[name = tensor<string, []>("add_2")];
+            tensor<fp32, [1, 11, ?]> mul_6 = mul(x = add_2, y = sub_1)[name = tensor<string, []>("mul_6")];
+            tensor<fp32, [1, 11, ?]> har_phase = add(x = mul_6, y = mul_3)[name = tensor<string, []>("har_phase")];
+            tensor<int32, []> var_137 = const()[name = tensor<string, []>("op_137"), val = tensor<int32, []>(1)];
+            tensor<bool, []> input_13_interleave_0 = const()[name = tensor<string, []>("input_13_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [1, 22, ?]> input_13 = concat(axis = var_137, interleave = input_13_interleave_0, values = (har_spec, har_phase))[name = tensor<string, []>("input_13")];
+            tensor<string, []> input_15_pad_type_0 = const()[name = tensor<string, []>("input_15_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_15_pad_0 = const()[name = tensor<string, []>("input_15_pad_0"), val = tensor<int32, [2]>([3, 3])];
+            tensor<int32, [1]> input_15_strides_0 = const()[name = tensor<string, []>("input_15_strides_0"), val = tensor<int32, [1]>([6])];
+            tensor<int32, [1]> input_15_dilations_0 = const()[name = tensor<string, []>("input_15_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_15_groups_0 = const()[name = tensor<string, []>("input_15_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> input_15 = conv(bias = noise_convs_0_bias, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = noise_convs_0_weight_palettized, x = input_13)[name = tensor<string, []>("input_15")];
+            tensor<fp32, []> var_160 = const()[name = tensor<string, []>("op_160"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<fp32, [1, 512]> h_1 = linear(bias = noise_res_0_adain1_0_fc_bias, weight = noise_res_0_adain1_0_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_1")];
+            tensor<int32, [3]> var_243 = const()[name = tensor<string, []>("op_243"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp32, [1, 512, 1]> h_3 = reshape(shape = var_243, x = h_1)[name = tensor<string, []>("h_3")];
+            tensor<int32, [2]> var_245_split_sizes_0 = const()[name = tensor<string, []>("op_245_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_245_axis_0 = const()[name = tensor<string, []>("op_245_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, 1]> var_245_0, tensor<fp32, [1, 256, 1]> var_245_1 = split(axis = var_245_axis_0, split_sizes = var_245_split_sizes_0, x = h_3)[name = tensor<string, []>("op_245")];
+            tensor<fp32, []> var_247_promoted = const()[name = tensor<string, []>("op_247_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 256, 1]> var_248 = add(x = var_245_0, y = var_247_promoted)[name = tensor<string, []>("op_248")];
+            tensor<fp32, [1, 256, ?]> var_251 = instance_norm(beta = noise_res_0_adain1_0_norm_bias, epsilon = var_160, gamma = noise_res_0_adain1_0_norm_weight, x = input_15)[name = tensor<string, []>("op_251")];
+            tensor<fp32, [1, 256, ?]> var_252 = mul(x = var_248, y = var_251)[name = tensor<string, []>("op_252")];
+            tensor<fp32, [1, 256, ?]> xt_1 = add(x = var_252, y = var_245_1)[name = tensor<string, []>("xt_1")];
+            tensor<fp32, [1, 256, 1]> var_254 = const()[name = tensor<string, []>("op_254"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(722496)))];
+            tensor<fp32, [1, 256, ?]> var_257 = mul(x = noise_res_0_alpha1_0, y = xt_1)[name = tensor<string, []>("op_257")];
+            tensor<fp32, [1, 256, ?]> var_258 = sin(x = var_257)[name = tensor<string, []>("op_258")];
+            tensor<fp32, []> var_159_promoted = const()[name = tensor<string, []>("op_159_promoted"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 256, ?]> var_259 = pow(x = var_258, y = var_159_promoted)[name = tensor<string, []>("op_259")];
+            tensor<fp32, [1, 256, ?]> var_260 = mul(x = var_254, y = var_259)[name = tensor<string, []>("op_260")];
+            tensor<fp32, [1, 256, ?]> input_17 = add(x = xt_1, y = var_260)[name = tensor<string, []>("input_17")];
+            tensor<fp32, [256, 256, 7]> weight_9_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [458752]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723584))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1182400))), name = tensor<string, []>("weight_9_palettized"), shape = tensor<uint32, [3]>([256, 256, 7])];
+            tensor<string, []> input_19_pad_type_0 = const()[name = tensor<string, []>("input_19_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_19_pad_0 = const()[name = tensor<string, []>("input_19_pad_0"), val = tensor<int32, [2]>([3, 3])];
+            tensor<int32, [1]> input_19_strides_0 = const()[name = tensor<string, []>("input_19_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_19_dilations_0 = const()[name = tensor<string, []>("input_19_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_19_groups_0 = const()[name = tensor<string, []>("input_19_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> input_19 = conv(bias = noise_res_0_convs1_0_bias, dilations = input_19_dilations_0, groups = input_19_groups_0, pad = input_19_pad_0, pad_type = input_19_pad_type_0, strides = input_19_strides_0, weight = weight_9_palettized, x = input_17)[name = tensor<string, []>("input_19")];
+            tensor<fp32, [1, 512]> h_5 = linear(bias = noise_res_0_adain2_0_fc_bias, weight = noise_res_0_adain2_0_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_2")];
+            tensor<int32, [3]> var_276 = const()[name = tensor<string, []>("op_276"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp32, [1, 512, 1]> h_7 = reshape(shape = var_276, x = h_5)[name = tensor<string, []>("h_7")];
+            tensor<int32, [2]> var_278_split_sizes_0 = const()[name = tensor<string, []>("op_278_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_278_axis_0 = const()[name = tensor<string, []>("op_278_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, 1]> var_278_0, tensor<fp32, [1, 256, 1]> var_278_1 = split(axis = var_278_axis_0, split_sizes = var_278_split_sizes_0, x = h_7)[name = tensor<string, []>("op_278")];
+            tensor<fp32, []> var_280_promoted = const()[name = tensor<string, []>("op_280_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 256, 1]> var_281 = add(x = var_278_0, y = var_280_promoted)[name = tensor<string, []>("op_281")];
+            tensor<fp32, [1, 256, ?]> var_284 = instance_norm(beta = noise_res_0_adain1_0_norm_bias, epsilon = var_160, gamma = noise_res_0_adain1_0_norm_weight, x = input_19)[name = tensor<string, []>("op_284")];
+            tensor<fp32, [1, 256, ?]> var_285 = mul(x = var_281, y = var_284)[name = tensor<string, []>("op_285")];
+            tensor<fp32, [1, 256, ?]> xt_3 = add(x = var_285, y = var_278_1)[name = tensor<string, []>("xt_3")];
+            tensor<fp32, [1, 256, 1]> var_287 = const()[name = tensor<string, []>("op_287"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183488)))];
+            tensor<fp32, [1, 256, ?]> var_290 = mul(x = noise_res_0_alpha2_0, y = xt_3)[name = tensor<string, []>("op_290")];
+            tensor<fp32, [1, 256, ?]> var_291 = sin(x = var_290)[name = tensor<string, []>("op_291")];
+            tensor<fp32, []> var_159_promoted_1 = const()[name = tensor<string, []>("op_159_promoted_1"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 256, ?]> var_292 = pow(x = var_291, y = var_159_promoted_1)[name = tensor<string, []>("op_292")];
+            tensor<fp32, [1, 256, ?]> var_293 = mul(x = var_287, y = var_292)[name = tensor<string, []>("op_293")];
+            tensor<fp32, [1, 256, ?]> input_21 = add(x = xt_3, y = var_293)[name = tensor<string, []>("input_21")];
+            tensor<fp32, [256, 256, 7]> weight_13_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [458752]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1184576))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1643392))), name = tensor<string, []>("weight_13_palettized"), shape = tensor<uint32, [3]>([256, 256, 7])];
+            tensor<string, []> xt_5_pad_type_0 = const()[name = tensor<string, []>("xt_5_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> xt_5_pad_0 = const()[name = tensor<string, []>("xt_5_pad_0"), val = tensor<int32, [2]>([3, 3])];
+            tensor<int32, [1]> xt_5_strides_0 = const()[name = tensor<string, []>("xt_5_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> xt_5_dilations_0 = const()[name = tensor<string, []>("xt_5_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> xt_5_groups_0 = const()[name = tensor<string, []>("xt_5_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> xt_5 = conv(bias = noise_res_0_convs2_0_bias, dilations = xt_5_dilations_0, groups = xt_5_groups_0, pad = xt_5_pad_0, pad_type = xt_5_pad_type_0, strides = xt_5_strides_0, weight = weight_13_palettized, x = input_21)[name = tensor<string, []>("xt_5")];
+            tensor<fp32, [1, 256, ?]> input_23 = add(x = xt_5, y = input_15)[name = tensor<string, []>("input_23")];
+            tensor<fp32, [1, 512]> h_9 = linear(bias = noise_res_0_adain1_1_fc_bias, weight = noise_res_0_adain1_1_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_3")];
+            tensor<int32, [3]> var_310 = const()[name = tensor<string, []>("op_310"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp32, [1, 512, 1]> h_11 = reshape(shape = var_310, x = h_9)[name = tensor<string, []>("h_11")];
+            tensor<int32, [2]> var_312_split_sizes_0 = const()[name = tensor<string, []>("op_312_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_312_axis_0 = const()[name = tensor<string, []>("op_312_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, 1]> var_312_0, tensor<fp32, [1, 256, 1]> var_312_1 = split(axis = var_312_axis_0, split_sizes = var_312_split_sizes_0, x = h_11)[name = tensor<string, []>("op_312")];
+            tensor<fp32, []> var_314_promoted = const()[name = tensor<string, []>("op_314_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 256, 1]> var_315 = add(x = var_312_0, y = var_314_promoted)[name = tensor<string, []>("op_315")];
+            tensor<fp32, [1, 256, ?]> var_318 = instance_norm(beta = noise_res_0_adain1_0_norm_bias, epsilon = var_160, gamma = noise_res_0_adain1_0_norm_weight, x = input_23)[name = tensor<string, []>("op_318")];
+            tensor<fp32, [1, 256, ?]> var_319 = mul(x = var_315, y = var_318)[name = tensor<string, []>("op_319")];
+            tensor<fp32, [1, 256, ?]> xt_7 = add(x = var_319, y = var_312_1)[name = tensor<string, []>("xt_7")];
+            tensor<fp32, [1, 256, 1]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1644480)))];
+            tensor<fp32, [1, 256, ?]> var_324 = mul(x = noise_res_0_alpha1_1, y = xt_7)[name = tensor<string, []>("op_324")];
+            tensor<fp32, [1, 256, ?]> var_325 = sin(x = var_324)[name = tensor<string, []>("op_325")];
+            tensor<fp32, []> var_159_promoted_2 = const()[name = tensor<string, []>("op_159_promoted_2"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 256, ?]> var_326 = pow(x = var_325, y = var_159_promoted_2)[name = tensor<string, []>("op_326")];
+            tensor<fp32, [1, 256, ?]> var_327 = mul(x = var_321, y = var_326)[name = tensor<string, []>("op_327")];
+            tensor<fp32, [1, 256, ?]> input_25 = add(x = xt_7, y = var_327)[name = tensor<string, []>("input_25")];
+            tensor<fp32, [256, 256, 7]> weight_17_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [458752]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645568))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2104384))), name = tensor<string, []>("weight_17_palettized"), shape = tensor<uint32, [3]>([256, 256, 7])];
+            tensor<string, []> input_27_pad_type_0 = const()[name = tensor<string, []>("input_27_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_27_pad_0 = const()[name = tensor<string, []>("input_27_pad_0"), val = tensor<int32, [2]>([9, 9])];
+            tensor<int32, [1]> input_27_dilations_0 = const()[name = tensor<string, []>("input_27_dilations_0"), val = tensor<int32, [1]>([3])];
+            tensor<int32, [1]> input_27_strides_0 = const()[name = tensor<string, []>("input_27_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_27_groups_0 = const()[name = tensor<string, []>("input_27_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> input_27 = conv(bias = noise_res_0_convs1_1_bias, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = weight_17_palettized, x = input_25)[name = tensor<string, []>("input_27")];
+            tensor<fp32, [1, 512]> h_13 = linear(bias = noise_res_0_adain2_1_fc_bias, weight = noise_res_0_adain2_1_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_4")];
+            tensor<int32, [3]> var_343 = const()[name = tensor<string, []>("op_343"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp32, [1, 512, 1]> h_15 = reshape(shape = var_343, x = h_13)[name = tensor<string, []>("h_15")];
+            tensor<int32, [2]> var_345_split_sizes_0 = const()[name = tensor<string, []>("op_345_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_345_axis_0 = const()[name = tensor<string, []>("op_345_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, 1]> var_345_0, tensor<fp32, [1, 256, 1]> var_345_1 = split(axis = var_345_axis_0, split_sizes = var_345_split_sizes_0, x = h_15)[name = tensor<string, []>("op_345")];
+            tensor<fp32, []> var_347_promoted = const()[name = tensor<string, []>("op_347_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 256, 1]> var_348 = add(x = var_345_0, y = var_347_promoted)[name = tensor<string, []>("op_348")];
+            tensor<fp32, [1, 256, ?]> var_351 = instance_norm(beta = noise_res_0_adain1_0_norm_bias, epsilon = var_160, gamma = noise_res_0_adain1_0_norm_weight, x = input_27)[name = tensor<string, []>("op_351")];
+            tensor<fp32, [1, 256, ?]> var_352 = mul(x = var_348, y = var_351)[name = tensor<string, []>("op_352")];
+            tensor<fp32, [1, 256, ?]> xt_9 = add(x = var_352, y = var_345_1)[name = tensor<string, []>("xt_9")];
+            tensor<fp32, [1, 256, 1]> var_354 = const()[name = tensor<string, []>("op_354"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2105472)))];
+            tensor<fp32, [1, 256, ?]> var_357 = mul(x = noise_res_0_alpha2_1, y = xt_9)[name = tensor<string, []>("op_357")];
+            tensor<fp32, [1, 256, ?]> var_358 = sin(x = var_357)[name = tensor<string, []>("op_358")];
+            tensor<fp32, []> var_159_promoted_3 = const()[name = tensor<string, []>("op_159_promoted_3"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 256, ?]> var_359 = pow(x = var_358, y = var_159_promoted_3)[name = tensor<string, []>("op_359")];
+            tensor<fp32, [1, 256, ?]> var_360 = mul(x = var_354, y = var_359)[name = tensor<string, []>("op_360")];
+            tensor<fp32, [1, 256, ?]> input_29 = add(x = xt_9, y = var_360)[name = tensor<string, []>("input_29")];
+            tensor<fp32, [256, 256, 7]> weight_21_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [458752]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2106560))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2565376))), name = tensor<string, []>("weight_21_palettized"), shape = tensor<uint32, [3]>([256, 256, 7])];
+            tensor<string, []> xt_11_pad_type_0 = const()[name = tensor<string, []>("xt_11_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> xt_11_pad_0 = const()[name = tensor<string, []>("xt_11_pad_0"), val = tensor<int32, [2]>([3, 3])];
+            tensor<int32, [1]> xt_11_strides_0 = const()[name = tensor<string, []>("xt_11_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> xt_11_dilations_0 = const()[name = tensor<string, []>("xt_11_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> xt_11_groups_0 = const()[name = tensor<string, []>("xt_11_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> xt_11 = conv(bias = noise_res_0_convs2_1_bias, dilations = xt_11_dilations_0, groups = xt_11_groups_0, pad = xt_11_pad_0, pad_type = xt_11_pad_type_0, strides = xt_11_strides_0, weight = weight_21_palettized, x = input_29)[name = tensor<string, []>("xt_11")];
+            tensor<fp32, [1, 256, ?]> input_31 = add(x = xt_11, y = input_23)[name = tensor<string, []>("input_31")];
+            tensor<fp32, [1, 512]> h_17 = linear(bias = noise_res_0_adain1_2_fc_bias, weight = noise_res_0_adain1_2_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_5")];
+            tensor<int32, [3]> var_377 = const()[name = tensor<string, []>("op_377"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp32, [1, 512, 1]> h_19 = reshape(shape = var_377, x = h_17)[name = tensor<string, []>("h_19")];
+            tensor<int32, [2]> var_379_split_sizes_0 = const()[name = tensor<string, []>("op_379_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_379_axis_0 = const()[name = tensor<string, []>("op_379_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, 1]> var_379_0, tensor<fp32, [1, 256, 1]> var_379_1 = split(axis = var_379_axis_0, split_sizes = var_379_split_sizes_0, x = h_19)[name = tensor<string, []>("op_379")];
+            tensor<fp32, []> var_381_promoted = const()[name = tensor<string, []>("op_381_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 256, 1]> var_382 = add(x = var_379_0, y = var_381_promoted)[name = tensor<string, []>("op_382")];
+            tensor<fp32, [1, 256, ?]> var_385 = instance_norm(beta = noise_res_0_adain1_0_norm_bias, epsilon = var_160, gamma = noise_res_0_adain1_0_norm_weight, x = input_31)[name = tensor<string, []>("op_385")];
+            tensor<fp32, [1, 256, ?]> var_386 = mul(x = var_382, y = var_385)[name = tensor<string, []>("op_386")];
+            tensor<fp32, [1, 256, ?]> xt_13 = add(x = var_386, y = var_379_1)[name = tensor<string, []>("xt_13")];
+            tensor<fp32, [1, 256, 1]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2566464)))];
+            tensor<fp32, [1, 256, ?]> var_391 = mul(x = noise_res_0_alpha1_2, y = xt_13)[name = tensor<string, []>("op_391")];
+            tensor<fp32, [1, 256, ?]> var_392 = sin(x = var_391)[name = tensor<string, []>("op_392")];
+            tensor<fp32, []> var_159_promoted_4 = const()[name = tensor<string, []>("op_159_promoted_4"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 256, ?]> var_393 = pow(x = var_392, y = var_159_promoted_4)[name = tensor<string, []>("op_393")];
+            tensor<fp32, [1, 256, ?]> var_394 = mul(x = var_388, y = var_393)[name = tensor<string, []>("op_394")];
+            tensor<fp32, [1, 256, ?]> input_33 = add(x = xt_13, y = var_394)[name = tensor<string, []>("input_33")];
+            tensor<fp32, [256, 256, 7]> weight_25_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [458752]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2567552))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3026368))), name = tensor<string, []>("weight_25_palettized"), shape = tensor<uint32, [3]>([256, 256, 7])];
+            tensor<string, []> input_35_pad_type_0 = const()[name = tensor<string, []>("input_35_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_35_pad_0 = const()[name = tensor<string, []>("input_35_pad_0"), val = tensor<int32, [2]>([15, 15])];
+            tensor<int32, [1]> input_35_dilations_0 = const()[name = tensor<string, []>("input_35_dilations_0"), val = tensor<int32, [1]>([5])];
+            tensor<int32, [1]> input_35_strides_0 = const()[name = tensor<string, []>("input_35_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_35_groups_0 = const()[name = tensor<string, []>("input_35_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> input_35 = conv(bias = noise_res_0_convs1_2_bias, dilations = input_35_dilations_0, groups = input_35_groups_0, pad = input_35_pad_0, pad_type = input_35_pad_type_0, strides = input_35_strides_0, weight = weight_25_palettized, x = input_33)[name = tensor<string, []>("input_35")];
+            tensor<fp32, [1, 512]> h_21 = linear(bias = noise_res_0_adain2_2_fc_bias, weight = noise_res_0_adain2_2_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_6")];
+            tensor<int32, [3]> var_410 = const()[name = tensor<string, []>("op_410"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp32, [1, 512, 1]> h_23 = reshape(shape = var_410, x = h_21)[name = tensor<string, []>("h_23")];
+            tensor<int32, [2]> var_412_split_sizes_0 = const()[name = tensor<string, []>("op_412_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_412_axis_0 = const()[name = tensor<string, []>("op_412_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, 1]> var_412_0, tensor<fp32, [1, 256, 1]> var_412_1 = split(axis = var_412_axis_0, split_sizes = var_412_split_sizes_0, x = h_23)[name = tensor<string, []>("op_412")];
+            tensor<fp32, []> var_414_promoted = const()[name = tensor<string, []>("op_414_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 256, 1]> var_415 = add(x = var_412_0, y = var_414_promoted)[name = tensor<string, []>("op_415")];
+            tensor<fp32, [1, 256, ?]> var_418 = instance_norm(beta = noise_res_0_adain1_0_norm_bias, epsilon = var_160, gamma = noise_res_0_adain1_0_norm_weight, x = input_35)[name = tensor<string, []>("op_418")];
+            tensor<fp32, [1, 256, ?]> var_419 = mul(x = var_415, y = var_418)[name = tensor<string, []>("op_419")];
+            tensor<fp32, [1, 256, ?]> xt_15 = add(x = var_419, y = var_412_1)[name = tensor<string, []>("xt_15")];
+            tensor<fp32, [1, 256, 1]> var_421 = const()[name = tensor<string, []>("op_421"), val = tensor<fp32, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3027456)))];
+            tensor<fp32, [1, 256, ?]> var_424 = mul(x = noise_res_0_alpha2_2, y = xt_15)[name = tensor<string, []>("op_424")];
+            tensor<fp32, [1, 256, ?]> var_425 = sin(x = var_424)[name = tensor<string, []>("op_425")];
+            tensor<fp32, []> var_159_promoted_5 = const()[name = tensor<string, []>("op_159_promoted_5"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 256, ?]> var_426 = pow(x = var_425, y = var_159_promoted_5)[name = tensor<string, []>("op_426")];
+            tensor<fp32, [1, 256, ?]> var_427 = mul(x = var_421, y = var_426)[name = tensor<string, []>("op_427")];
+            tensor<fp32, [1, 256, ?]> input_37 = add(x = xt_15, y = var_427)[name = tensor<string, []>("input_37")];
+            tensor<fp32, [256, 256, 7]> weight_29_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [458752]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3028544))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3487360))), name = tensor<string, []>("weight_29_palettized"), shape = tensor<uint32, [3]>([256, 256, 7])];
+            tensor<string, []> xt_17_pad_type_0 = const()[name = tensor<string, []>("xt_17_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> xt_17_pad_0 = const()[name = tensor<string, []>("xt_17_pad_0"), val = tensor<int32, [2]>([3, 3])];
+            tensor<int32, [1]> xt_17_strides_0 = const()[name = tensor<string, []>("xt_17_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> xt_17_dilations_0 = const()[name = tensor<string, []>("xt_17_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> xt_17_groups_0 = const()[name = tensor<string, []>("xt_17_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 256, ?]> xt_17 = conv(bias = noise_res_0_convs2_2_bias, dilations = xt_17_dilations_0, groups = xt_17_groups_0, pad = xt_17_pad_0, pad_type = xt_17_pad_type_0, strides = xt_17_strides_0, weight = weight_29_palettized, x = input_37)[name = tensor<string, []>("xt_17")];
+            tensor<fp32, [1, 256, ?]> x_source_0 = add(x = xt_17, y = input_31)[name = tensor<string, []>("op_436")];
+            tensor<string, []> input_39_pad_type_0 = const()[name = tensor<string, []>("input_39_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> input_39_strides_0 = const()[name = tensor<string, []>("input_39_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_39_pad_0 = const()[name = tensor<string, []>("input_39_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_39_dilations_0 = const()[name = tensor<string, []>("input_39_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_39_groups_0 = const()[name = tensor<string, []>("input_39_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> input_39 = conv(bias = noise_convs_1_bias, dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = noise_convs_1_weight_palettized, x = input_13)[name = tensor<string, []>("input_39")];
+            tensor<fp32, []> var_456 = const()[name = tensor<string, []>("op_456"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<fp32, [1, 256]> h_25 = linear(bias = noise_res_1_adain1_0_fc_bias, weight = noise_res_1_adain1_0_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_7")];
+            tensor<int32, [3]> var_539 = const()[name = tensor<string, []>("op_539"), val = tensor<int32, [3]>([1, 256, 1])];
+            tensor<fp32, [1, 256, 1]> h_27 = reshape(shape = var_539, x = h_25)[name = tensor<string, []>("h_27")];
+            tensor<int32, [2]> var_541_split_sizes_0 = const()[name = tensor<string, []>("op_541_split_sizes_0"), val = tensor<int32, [2]>([128, 128])];
+            tensor<int32, []> var_541_axis_0 = const()[name = tensor<string, []>("op_541_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, 1]> var_541_0, tensor<fp32, [1, 128, 1]> var_541_1 = split(axis = var_541_axis_0, split_sizes = var_541_split_sizes_0, x = h_27)[name = tensor<string, []>("op_541")];
+            tensor<fp32, []> var_543_promoted = const()[name = tensor<string, []>("op_543_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 128, 1]> var_544 = add(x = var_541_0, y = var_543_promoted)[name = tensor<string, []>("op_544")];
+            tensor<fp32, [1, 128, ?]> var_547 = instance_norm(beta = noise_res_1_adain1_0_norm_bias, epsilon = var_456, gamma = noise_res_1_adain1_0_norm_weight, x = input_39)[name = tensor<string, []>("op_547")];
+            tensor<fp32, [1, 128, ?]> var_548 = mul(x = var_544, y = var_547)[name = tensor<string, []>("op_548")];
+            tensor<fp32, [1, 128, ?]> xt_19 = add(x = var_548, y = var_541_1)[name = tensor<string, []>("xt_19")];
+            tensor<fp32, [1, 128, 1]> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3488448)))];
+            tensor<fp32, [1, 128, ?]> var_553 = mul(x = noise_res_1_alpha1_0, y = xt_19)[name = tensor<string, []>("op_553")];
+            tensor<fp32, [1, 128, ?]> var_554 = sin(x = var_553)[name = tensor<string, []>("op_554")];
+            tensor<fp32, []> var_455_promoted = const()[name = tensor<string, []>("op_455_promoted"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 128, ?]> var_555 = pow(x = var_554, y = var_455_promoted)[name = tensor<string, []>("op_555")];
+            tensor<fp32, [1, 128, ?]> var_556 = mul(x = var_550, y = var_555)[name = tensor<string, []>("op_556")];
+            tensor<fp32, [1, 128, ?]> input_41 = add(x = xt_19, y = var_556)[name = tensor<string, []>("input_41")];
+            tensor<fp32, [128, 128, 11]> weight_35_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [180224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3489024))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3669312))), name = tensor<string, []>("weight_35_palettized"), shape = tensor<uint32, [3]>([128, 128, 11])];
+            tensor<string, []> input_43_pad_type_0 = const()[name = tensor<string, []>("input_43_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_43_pad_0 = const()[name = tensor<string, []>("input_43_pad_0"), val = tensor<int32, [2]>([5, 5])];
+            tensor<int32, [1]> input_43_strides_0 = const()[name = tensor<string, []>("input_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_43_dilations_0 = const()[name = tensor<string, []>("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_43_groups_0 = const()[name = tensor<string, []>("input_43_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> input_43 = conv(bias = noise_res_1_convs1_0_bias, dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = weight_35_palettized, x = input_41)[name = tensor<string, []>("input_43")];
+            tensor<fp32, [1, 256]> h_29 = linear(bias = noise_res_1_adain2_0_fc_bias, weight = noise_res_1_adain2_0_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_8")];
+            tensor<int32, [3]> var_572 = const()[name = tensor<string, []>("op_572"), val = tensor<int32, [3]>([1, 256, 1])];
+            tensor<fp32, [1, 256, 1]> h_31 = reshape(shape = var_572, x = h_29)[name = tensor<string, []>("h_31")];
+            tensor<int32, [2]> var_574_split_sizes_0 = const()[name = tensor<string, []>("op_574_split_sizes_0"), val = tensor<int32, [2]>([128, 128])];
+            tensor<int32, []> var_574_axis_0 = const()[name = tensor<string, []>("op_574_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, 1]> var_574_0, tensor<fp32, [1, 128, 1]> var_574_1 = split(axis = var_574_axis_0, split_sizes = var_574_split_sizes_0, x = h_31)[name = tensor<string, []>("op_574")];
+            tensor<fp32, []> var_576_promoted = const()[name = tensor<string, []>("op_576_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 128, 1]> var_577 = add(x = var_574_0, y = var_576_promoted)[name = tensor<string, []>("op_577")];
+            tensor<fp32, [1, 128, ?]> var_580 = instance_norm(beta = noise_res_1_adain1_0_norm_bias, epsilon = var_456, gamma = noise_res_1_adain1_0_norm_weight, x = input_43)[name = tensor<string, []>("op_580")];
+            tensor<fp32, [1, 128, ?]> var_581 = mul(x = var_577, y = var_580)[name = tensor<string, []>("op_581")];
+            tensor<fp32, [1, 128, ?]> xt_21 = add(x = var_581, y = var_574_1)[name = tensor<string, []>("xt_21")];
+            tensor<fp32, [1, 128, 1]> var_583 = const()[name = tensor<string, []>("op_583"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3670400)))];
+            tensor<fp32, [1, 128, ?]> var_586 = mul(x = noise_res_1_alpha2_0, y = xt_21)[name = tensor<string, []>("op_586")];
+            tensor<fp32, [1, 128, ?]> var_587 = sin(x = var_586)[name = tensor<string, []>("op_587")];
+            tensor<fp32, []> var_455_promoted_1 = const()[name = tensor<string, []>("op_455_promoted_1"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 128, ?]> var_588 = pow(x = var_587, y = var_455_promoted_1)[name = tensor<string, []>("op_588")];
+            tensor<fp32, [1, 128, ?]> var_589 = mul(x = var_583, y = var_588)[name = tensor<string, []>("op_589")];
+            tensor<fp32, [1, 128, ?]> input_45 = add(x = xt_21, y = var_589)[name = tensor<string, []>("input_45")];
+            tensor<fp32, [128, 128, 11]> weight_39_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [180224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3670976))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3851264))), name = tensor<string, []>("weight_39_palettized"), shape = tensor<uint32, [3]>([128, 128, 11])];
+            tensor<string, []> xt_23_pad_type_0 = const()[name = tensor<string, []>("xt_23_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> xt_23_pad_0 = const()[name = tensor<string, []>("xt_23_pad_0"), val = tensor<int32, [2]>([5, 5])];
+            tensor<int32, [1]> xt_23_strides_0 = const()[name = tensor<string, []>("xt_23_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> xt_23_dilations_0 = const()[name = tensor<string, []>("xt_23_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> xt_23_groups_0 = const()[name = tensor<string, []>("xt_23_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> xt_23 = conv(bias = noise_res_1_convs2_0_bias, dilations = xt_23_dilations_0, groups = xt_23_groups_0, pad = xt_23_pad_0, pad_type = xt_23_pad_type_0, strides = xt_23_strides_0, weight = weight_39_palettized, x = input_45)[name = tensor<string, []>("xt_23")];
+            tensor<fp32, [1, 128, ?]> input_47 = add(x = xt_23, y = input_39)[name = tensor<string, []>("input_47")];
+            tensor<fp32, [1, 256]> h_33 = linear(bias = noise_res_1_adain1_1_fc_bias, weight = noise_res_1_adain1_1_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_9")];
+            tensor<int32, [3]> var_606 = const()[name = tensor<string, []>("op_606"), val = tensor<int32, [3]>([1, 256, 1])];
+            tensor<fp32, [1, 256, 1]> h_35 = reshape(shape = var_606, x = h_33)[name = tensor<string, []>("h_35")];
+            tensor<int32, [2]> var_608_split_sizes_0 = const()[name = tensor<string, []>("op_608_split_sizes_0"), val = tensor<int32, [2]>([128, 128])];
+            tensor<int32, []> var_608_axis_0 = const()[name = tensor<string, []>("op_608_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, 1]> var_608_0, tensor<fp32, [1, 128, 1]> var_608_1 = split(axis = var_608_axis_0, split_sizes = var_608_split_sizes_0, x = h_35)[name = tensor<string, []>("op_608")];
+            tensor<fp32, []> var_610_promoted = const()[name = tensor<string, []>("op_610_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 128, 1]> var_611 = add(x = var_608_0, y = var_610_promoted)[name = tensor<string, []>("op_611")];
+            tensor<fp32, [1, 128, ?]> var_614 = instance_norm(beta = noise_res_1_adain1_0_norm_bias, epsilon = var_456, gamma = noise_res_1_adain1_0_norm_weight, x = input_47)[name = tensor<string, []>("op_614")];
+            tensor<fp32, [1, 128, ?]> var_615 = mul(x = var_611, y = var_614)[name = tensor<string, []>("op_615")];
+            tensor<fp32, [1, 128, ?]> xt_25 = add(x = var_615, y = var_608_1)[name = tensor<string, []>("xt_25")];
+            tensor<fp32, [1, 128, 1]> var_617 = const()[name = tensor<string, []>("op_617"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3852352)))];
+            tensor<fp32, [1, 128, ?]> var_620 = mul(x = noise_res_1_alpha1_1, y = xt_25)[name = tensor<string, []>("op_620")];
+            tensor<fp32, [1, 128, ?]> var_621 = sin(x = var_620)[name = tensor<string, []>("op_621")];
+            tensor<fp32, []> var_455_promoted_2 = const()[name = tensor<string, []>("op_455_promoted_2"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 128, ?]> var_622 = pow(x = var_621, y = var_455_promoted_2)[name = tensor<string, []>("op_622")];
+            tensor<fp32, [1, 128, ?]> var_623 = mul(x = var_617, y = var_622)[name = tensor<string, []>("op_623")];
+            tensor<fp32, [1, 128, ?]> input_49 = add(x = xt_25, y = var_623)[name = tensor<string, []>("input_49")];
+            tensor<fp32, [128, 128, 11]> weight_43_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [180224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3852928))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4033216))), name = tensor<string, []>("weight_43_palettized"), shape = tensor<uint32, [3]>([128, 128, 11])];
+            tensor<string, []> input_51_pad_type_0 = const()[name = tensor<string, []>("input_51_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_51_pad_0 = const()[name = tensor<string, []>("input_51_pad_0"), val = tensor<int32, [2]>([15, 15])];
+            tensor<int32, [1]> input_51_dilations_0 = const()[name = tensor<string, []>("input_51_dilations_0"), val = tensor<int32, [1]>([3])];
+            tensor<int32, [1]> input_51_strides_0 = const()[name = tensor<string, []>("input_51_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_51_groups_0 = const()[name = tensor<string, []>("input_51_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> input_51 = conv(bias = noise_res_1_convs1_1_bias, dilations = input_51_dilations_0, groups = input_51_groups_0, pad = input_51_pad_0, pad_type = input_51_pad_type_0, strides = input_51_strides_0, weight = weight_43_palettized, x = input_49)[name = tensor<string, []>("input_51")];
+            tensor<fp32, [1, 256]> h_37 = linear(bias = noise_res_1_adain2_1_fc_bias, weight = noise_res_1_adain2_1_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_10")];
+            tensor<int32, [3]> var_639 = const()[name = tensor<string, []>("op_639"), val = tensor<int32, [3]>([1, 256, 1])];
+            tensor<fp32, [1, 256, 1]> h_39 = reshape(shape = var_639, x = h_37)[name = tensor<string, []>("h_39")];
+            tensor<int32, [2]> var_641_split_sizes_0 = const()[name = tensor<string, []>("op_641_split_sizes_0"), val = tensor<int32, [2]>([128, 128])];
+            tensor<int32, []> var_641_axis_0 = const()[name = tensor<string, []>("op_641_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, 1]> var_641_0, tensor<fp32, [1, 128, 1]> var_641_1 = split(axis = var_641_axis_0, split_sizes = var_641_split_sizes_0, x = h_39)[name = tensor<string, []>("op_641")];
+            tensor<fp32, []> var_643_promoted = const()[name = tensor<string, []>("op_643_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 128, 1]> var_644 = add(x = var_641_0, y = var_643_promoted)[name = tensor<string, []>("op_644")];
+            tensor<fp32, [1, 128, ?]> var_647 = instance_norm(beta = noise_res_1_adain1_0_norm_bias, epsilon = var_456, gamma = noise_res_1_adain1_0_norm_weight, x = input_51)[name = tensor<string, []>("op_647")];
+            tensor<fp32, [1, 128, ?]> var_648 = mul(x = var_644, y = var_647)[name = tensor<string, []>("op_648")];
+            tensor<fp32, [1, 128, ?]> xt_27 = add(x = var_648, y = var_641_1)[name = tensor<string, []>("xt_27")];
+            tensor<fp32, [1, 128, 1]> var_650 = const()[name = tensor<string, []>("op_650"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4034304)))];
+            tensor<fp32, [1, 128, ?]> var_653 = mul(x = noise_res_1_alpha2_1, y = xt_27)[name = tensor<string, []>("op_653")];
+            tensor<fp32, [1, 128, ?]> var_654 = sin(x = var_653)[name = tensor<string, []>("op_654")];
+            tensor<fp32, []> var_455_promoted_3 = const()[name = tensor<string, []>("op_455_promoted_3"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 128, ?]> var_655 = pow(x = var_654, y = var_455_promoted_3)[name = tensor<string, []>("op_655")];
+            tensor<fp32, [1, 128, ?]> var_656 = mul(x = var_650, y = var_655)[name = tensor<string, []>("op_656")];
+            tensor<fp32, [1, 128, ?]> input_53 = add(x = xt_27, y = var_656)[name = tensor<string, []>("input_53")];
+            tensor<fp32, [128, 128, 11]> weight_47_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [180224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4034880))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4215168))), name = tensor<string, []>("weight_47_palettized"), shape = tensor<uint32, [3]>([128, 128, 11])];
+            tensor<string, []> xt_29_pad_type_0 = const()[name = tensor<string, []>("xt_29_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> xt_29_pad_0 = const()[name = tensor<string, []>("xt_29_pad_0"), val = tensor<int32, [2]>([5, 5])];
+            tensor<int32, [1]> xt_29_strides_0 = const()[name = tensor<string, []>("xt_29_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> xt_29_dilations_0 = const()[name = tensor<string, []>("xt_29_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> xt_29_groups_0 = const()[name = tensor<string, []>("xt_29_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> xt_29 = conv(bias = noise_res_1_convs2_1_bias, dilations = xt_29_dilations_0, groups = xt_29_groups_0, pad = xt_29_pad_0, pad_type = xt_29_pad_type_0, strides = xt_29_strides_0, weight = weight_47_palettized, x = input_53)[name = tensor<string, []>("xt_29")];
+            tensor<fp32, [1, 128, ?]> input_55 = add(x = xt_29, y = input_47)[name = tensor<string, []>("input_55")];
+            tensor<fp32, [1, 256]> h_41 = linear(bias = noise_res_1_adain1_2_fc_bias, weight = noise_res_1_adain1_2_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_11")];
+            tensor<int32, [3]> var_673 = const()[name = tensor<string, []>("op_673"), val = tensor<int32, [3]>([1, 256, 1])];
+            tensor<fp32, [1, 256, 1]> h_43 = reshape(shape = var_673, x = h_41)[name = tensor<string, []>("h_43")];
+            tensor<int32, [2]> var_675_split_sizes_0 = const()[name = tensor<string, []>("op_675_split_sizes_0"), val = tensor<int32, [2]>([128, 128])];
+            tensor<int32, []> var_675_axis_0 = const()[name = tensor<string, []>("op_675_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, 1]> var_675_0, tensor<fp32, [1, 128, 1]> var_675_1 = split(axis = var_675_axis_0, split_sizes = var_675_split_sizes_0, x = h_43)[name = tensor<string, []>("op_675")];
+            tensor<fp32, []> var_677_promoted = const()[name = tensor<string, []>("op_677_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 128, 1]> var_678 = add(x = var_675_0, y = var_677_promoted)[name = tensor<string, []>("op_678")];
+            tensor<fp32, [1, 128, ?]> var_681 = instance_norm(beta = noise_res_1_adain1_0_norm_bias, epsilon = var_456, gamma = noise_res_1_adain1_0_norm_weight, x = input_55)[name = tensor<string, []>("op_681")];
+            tensor<fp32, [1, 128, ?]> var_682 = mul(x = var_678, y = var_681)[name = tensor<string, []>("op_682")];
+            tensor<fp32, [1, 128, ?]> xt_31 = add(x = var_682, y = var_675_1)[name = tensor<string, []>("xt_31")];
+            tensor<fp32, [1, 128, 1]> var_684 = const()[name = tensor<string, []>("op_684"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4216256)))];
+            tensor<fp32, [1, 128, ?]> var_687 = mul(x = noise_res_1_alpha1_2, y = xt_31)[name = tensor<string, []>("op_687")];
+            tensor<fp32, [1, 128, ?]> var_688 = sin(x = var_687)[name = tensor<string, []>("op_688")];
+            tensor<fp32, []> var_455_promoted_4 = const()[name = tensor<string, []>("op_455_promoted_4"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 128, ?]> var_689 = pow(x = var_688, y = var_455_promoted_4)[name = tensor<string, []>("op_689")];
+            tensor<fp32, [1, 128, ?]> var_690 = mul(x = var_684, y = var_689)[name = tensor<string, []>("op_690")];
+            tensor<fp32, [1, 128, ?]> input_57 = add(x = xt_31, y = var_690)[name = tensor<string, []>("input_57")];
+            tensor<fp32, [128, 128, 11]> weight_51_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [180224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4216832))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4397120))), name = tensor<string, []>("weight_51_palettized"), shape = tensor<uint32, [3]>([128, 128, 11])];
+            tensor<string, []> input_59_pad_type_0 = const()[name = tensor<string, []>("input_59_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_59_pad_0 = const()[name = tensor<string, []>("input_59_pad_0"), val = tensor<int32, [2]>([25, 25])];
+            tensor<int32, [1]> input_59_dilations_0 = const()[name = tensor<string, []>("input_59_dilations_0"), val = tensor<int32, [1]>([5])];
+            tensor<int32, [1]> input_59_strides_0 = const()[name = tensor<string, []>("input_59_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_59_groups_0 = const()[name = tensor<string, []>("input_59_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> input_59 = conv(bias = noise_res_1_convs1_2_bias, dilations = input_59_dilations_0, groups = input_59_groups_0, pad = input_59_pad_0, pad_type = input_59_pad_type_0, strides = input_59_strides_0, weight = weight_51_palettized, x = input_57)[name = tensor<string, []>("input_59")];
+            tensor<fp32, [1, 256]> h_45 = linear(bias = noise_res_1_adain2_2_fc_bias, weight = noise_res_1_adain2_2_fc_weight_palettized, x = style_timbre)[name = tensor<string, []>("linear_12")];
+            tensor<int32, [3]> var_706 = const()[name = tensor<string, []>("op_706"), val = tensor<int32, [3]>([1, 256, 1])];
+            tensor<fp32, [1, 256, 1]> h = reshape(shape = var_706, x = h_45)[name = tensor<string, []>("h")];
+            tensor<int32, [2]> var_708_split_sizes_0 = const()[name = tensor<string, []>("op_708_split_sizes_0"), val = tensor<int32, [2]>([128, 128])];
+            tensor<int32, []> var_708_axis_0 = const()[name = tensor<string, []>("op_708_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, 1]> var_708_0, tensor<fp32, [1, 128, 1]> var_708_1 = split(axis = var_708_axis_0, split_sizes = var_708_split_sizes_0, x = h)[name = tensor<string, []>("op_708")];
+            tensor<fp32, []> var_710_promoted = const()[name = tensor<string, []>("op_710_promoted"), val = tensor<fp32, []>(0x1p+0)];
+            tensor<fp32, [1, 128, 1]> var_711 = add(x = var_708_0, y = var_710_promoted)[name = tensor<string, []>("op_711")];
+            tensor<fp32, [1, 128, ?]> var_714 = instance_norm(beta = noise_res_1_adain1_0_norm_bias, epsilon = var_456, gamma = noise_res_1_adain1_0_norm_weight, x = input_59)[name = tensor<string, []>("op_714")];
+            tensor<fp32, [1, 128, ?]> var_715 = mul(x = var_711, y = var_714)[name = tensor<string, []>("op_715")];
+            tensor<fp32, [1, 128, ?]> xt_33 = add(x = var_715, y = var_708_1)[name = tensor<string, []>("xt_33")];
+            tensor<fp32, [1, 128, 1]> var_717 = const()[name = tensor<string, []>("op_717"), val = tensor<fp32, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4398208)))];
+            tensor<fp32, [1, 128, ?]> var_720 = mul(x = noise_res_1_alpha2_2, y = xt_33)[name = tensor<string, []>("op_720")];
+            tensor<fp32, [1, 128, ?]> var_721 = sin(x = var_720)[name = tensor<string, []>("op_721")];
+            tensor<fp32, []> var_455_promoted_5 = const()[name = tensor<string, []>("op_455_promoted_5"), val = tensor<fp32, []>(0x1p+1)];
+            tensor<fp32, [1, 128, ?]> var_722 = pow(x = var_721, y = var_455_promoted_5)[name = tensor<string, []>("op_722")];
+            tensor<fp32, [1, 128, ?]> var_723 = mul(x = var_717, y = var_722)[name = tensor<string, []>("op_723")];
+            tensor<fp32, [1, 128, ?]> input = add(x = xt_33, y = var_723)[name = tensor<string, []>("input")];
+            tensor<fp32, [128, 128, 11]> weight_55_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [180224]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4398784))), lut = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4579072))), name = tensor<string, []>("weight_55_palettized"), shape = tensor<uint32, [3]>([128, 128, 11])];
+            tensor<string, []> xt_pad_type_0 = const()[name = tensor<string, []>("xt_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> xt_pad_0 = const()[name = tensor<string, []>("xt_pad_0"), val = tensor<int32, [2]>([5, 5])];
+            tensor<int32, [1]> xt_strides_0 = const()[name = tensor<string, []>("xt_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> xt_dilations_0 = const()[name = tensor<string, []>("xt_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> xt_groups_0 = const()[name = tensor<string, []>("xt_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 128, ?]> xt = conv(bias = noise_res_1_convs2_2_bias, dilations = xt_dilations_0, groups = xt_groups_0, pad = xt_pad_0, pad_type = xt_pad_type_0, strides = xt_strides_0, weight = weight_55_palettized, x = input)[name = tensor<string, []>("xt")];
+            tensor<fp32, [1, 128, ?]> x_source_1 = add(x = xt, y = input_55)[name = tensor<string, []>("op_732")];
+        } -> (x_source_0, x_source_1);
+}

ANE/ANE-zh/KokoroNoise.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1a1edc7ac86b9899cb908d6e4e65f3def26ee7720d8730444bee29385596ab5
+size 4580160

ANE/ANE-zh/KokoroNoise.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8674162d48113407da93aeee21d7c2433738494bd3662094ba0249b8d0445e9b
+size 74960

ANE/ANE-zh/KokoroNoise.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1a1edc7ac86b9899cb908d6e4e65f3def26ee7720d8730444bee29385596ab5
+size 4580160

ANE/ANE-zh/KokoroNoise.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "510CA943-DA71-4F61-A81C-5534318B48C9": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "E69B7771-1977-4192-A042-A8C3807DC461": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "E69B7771-1977-4192-A042-A8C3807DC461"
+}

ANE/ANE-zh/KokoroPostAlbert.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ac50dad158a0dc349a240675d84ad920ae3202f282fc70fb2b833d78eb6829f
+size 243

ANE/ANE-zh/KokoroPostAlbert.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9246412170c6118149ccf412efafbb76db5bd47946776a1897b046ca2576f95
+size 556

ANE/ANE-zh/KokoroPostAlbert.mlmodelc/metadata.json

@@ -0,0 +1,148 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Mixed (Float16, Palettized (8 bits))",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "duration",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "d",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "t_en",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.equal" : 2,
+      "Ios17.reshape" : 3,
+      "Ios17.transpose" : 30,
+      "Split" : 3,
+      "Select" : 10,
+      "Ios17.expandDims" : 3,
+      "Ios17.add" : 6,
+      "Tile" : 1,
+      "Ios16.sigmoid" : 1,
+      "Ios16.reduceSum" : 1,
+      "Shape" : 1,
+      "Ios17.gather" : 2,
+      "Ios17.layerNorm" : 6,
+      "Ios17.lstm" : 5,
+      "Ios17.cast" : 3,
+      "Ios16.constexprLutToDense" : 29,
+      "Ios17.conv" : 3,
+      "Ios17.realDiv" : 2,
+      "Ios17.linear" : 5,
+      "Ios17.concat" : 5,
+      "Ios17.leakyRelu" : 3,
+      "Ios17.mul" : 3
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int16, Int32, UInt16)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-05-03",
+      "com.github.apple.coremltools.source" : "torch==2.11.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float16",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512 × 768",
+        "shapeRange" : "[[1, 1], [2, 512], [768, 768]]",
+        "formattedType" : "MultiArray (Float16 1 × 37 × 768)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37, 768]",
+        "name" : "bert_dur",
+        "shortDescription" : ""
+      },
+      {
+        "dataType" : "Int32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512",
+        "shapeRange" : "[[1, 1], [2, 512]]",
+        "formattedType" : "MultiArray (Int32 1 × 37)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37]",
+        "name" : "input_ids",
+        "shortDescription" : ""
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 128]",
+        "name" : "style_s",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "speed",
+        "type" : "MultiArray"
+      },
+      {
+        "dataType" : "Int32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 2...512",
+        "shapeRange" : "[[1, 1], [2, 512]]",
+        "formattedType" : "MultiArray (Int32 1 × 37)",
+        "type" : "MultiArray",
+        "shape" : "[1, 37]",
+        "name" : "attention_mask",
+        "shortDescription" : ""
+      }
+    ],
+    "generatedClassName" : "KokoroPostAlbert",
+    "method" : "predict"
+  }
+]

ANE/ANE-zh/KokoroPostAlbert.mlmodelc/model.mil

@@ -0,0 +1,277 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
+{
+    func main<ios17>(tensor<int32, [1, ?]> attention_mask, tensor<fp16, [1, ?, 768]> bert_dur, tensor<int32, [1, ?]> input_ids, tensor<fp16, [1]> speed, tensor<fp16, [1, 128]> style_s) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"attention_mask", [1, 37]}, {"bert_dur", [1, 37, 768]}, {"input_ids", [1, 37]}}), ("RangeDims", {{"attention_mask", [[1, 1], [2, 512]]}, {"bert_dur", [[1, 1], [2, 512], [768, 768]]}, {"input_ids", [[1, 1], [2, 512]]}})))] {
+            tensor<int32, []> var_11 = const()[name = tensor<string, []>("op_11"), val = tensor<int32, []>(0)];
+            tensor<bool, [1, ?]> m = equal(x = attention_mask, y = var_11)[name = tensor<string, []>("m")];
+            tensor<fp16, [512, 768]> bert_encoder_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [393216]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393344))), name = tensor<string, []>("bert_encoder_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 768])];
+            tensor<fp16, [512]> bert_encoder_bias_to_fp16 = const()[name = tensor<string, []>("bert_encoder_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393920)))];
+            tensor<fp16, [1, ?, 512]> linear_0_cast_fp16 = linear(bias = bert_encoder_bias_to_fp16, weight = bert_encoder_weight_to_fp16_palettized, x = bert_dur)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<int32, [3]> transpose_6_perm_0 = const()[name = tensor<string, []>("transpose_6_perm_0"), val = tensor<int32, [3]>([-2, 0, -1])];
+            tensor<int32, []> var_30 = const()[name = tensor<string, []>("op_30"), val = tensor<int32, []>(-1)];
+            tensor<int32, []> var_31 = const()[name = tensor<string, []>("op_31"), val = tensor<int32, []>(1)];
+            tensor<fp16, [?, 1, 512]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = linear_0_cast_fp16)[name = tensor<string, []>("transpose_47")];
+            tensor<int32, [3]> var_48_shape_cast_fp16 = shape(x = transpose_6_cast_fp16)[name = tensor<string, []>("op_48_shape_cast_fp16")];
+            tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> gather_0_validate_indices_0 = const()[name = tensor<string, []>("gather_0_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_48_shape_cast_fp16_to_int16_dtype_0 = const()[name = tensor<string, []>("op_48_shape_cast_fp16_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<uint16, []> gather_0_indices_0_to_uint16 = const()[name = tensor<string, []>("gather_0_indices_0_to_uint16"), val = tensor<uint16, []>(0)];
+            tensor<int16, [3]> var_48_shape_cast_fp16_to_int16 = cast(dtype = var_48_shape_cast_fp16_to_int16_dtype_0, x = var_48_shape_cast_fp16)[name = tensor<string, []>("cast_2")];
+            tensor<int16, []> gather_0_cast_uint16 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = gather_0_indices_0_to_uint16, validate_indices = gather_0_validate_indices_0, x = var_48_shape_cast_fp16_to_int16)[name = tensor<string, []>("gather_0_cast_uint16")];
+            tensor<string, []> gather_0_cast_uint16_to_int32_dtype_0 = const()[name = tensor<string, []>("gather_0_cast_uint16_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1]> var_49_axes_0 = const()[name = tensor<string, []>("op_49_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 128]> var_49_cast_fp16 = expand_dims(axes = var_49_axes_0, x = style_s)[name = tensor<string, []>("op_49_cast_fp16")];
+            tensor<int32, []> concat_0_axis_0 = const()[name = tensor<string, []>("concat_0_axis_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> concat_0_interleave_0 = const()[name = tensor<string, []>("concat_0_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, []> gather_0_cast_uint16_to_int32 = cast(dtype = gather_0_cast_uint16_to_int32_dtype_0, x = gather_0_cast_uint16)[name = tensor<string, []>("cast_1")];
+            tensor<int32, [3]> concat_0 = concat(axis = concat_0_axis_0, interleave = concat_0_interleave_0, values = (gather_0_cast_uint16_to_int32, var_30, var_30))[name = tensor<string, []>("concat_0")];
+            tensor<int32, [3]> shape_0 = const()[name = tensor<string, []>("shape_0"), val = tensor<int32, [3]>([1, 1, 128])];
+            tensor<int32, []> equal_0_y_0 = const()[name = tensor<string, []>("equal_0_y_0"), val = tensor<int32, []>(-1)];
+            tensor<bool, [3]> equal_0 = equal(x = concat_0, y = equal_0_y_0)[name = tensor<string, []>("equal_0")];
+            tensor<int32, [3]> select_0 = select(a = shape_0, b = concat_0, cond = equal_0)[name = tensor<string, []>("select_0")];
+            tensor<int32, [3]> real_div_0 = real_div(x = select_0, y = shape_0)[name = tensor<string, []>("real_div_0")];
+            tensor<fp16, [?, ?, ?]> s_cast_fp16 = tile(reps = real_div_0, x = var_49_cast_fp16)[name = tensor<string, []>("s_cast_fp16")];
+            tensor<bool, []> x_5_interleave_0 = const()[name = tensor<string, []>("x_5_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [?, 1, ?]> x_5_cast_fp16 = concat(axis = var_30, interleave = x_5_interleave_0, values = (transpose_6_cast_fp16, s_cast_fp16))[name = tensor<string, []>("x_5_cast_fp16")];
+            tensor<int32, [1]> var_54_axes_0 = const()[name = tensor<string, []>("op_54_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, [1, ?, 1]> var_54 = expand_dims(axes = var_54_axes_0, x = m)[name = tensor<string, []>("op_54")];
+            tensor<int32, [3]> var_55_perm_0 = const()[name = tensor<string, []>("op_55_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp16, []> var_28_to_fp16 = const()[name = tensor<string, []>("op_28_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<bool, [?, 1, 1]> var_55 = transpose(perm = var_55_perm_0, x = var_54)[name = tensor<string, []>("transpose_46")];
+            tensor<fp16, [?, 1, ?]> x_7_cast_fp16 = select(a = var_28_to_fp16, b = x_5_cast_fp16, cond = var_55)[name = tensor<string, []>("x_7_cast_fp16")];
+            tensor<string, []> input_3_batch_first_direction_0 = const()[name = tensor<string, []>("input_3_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> input_3_batch_first_output_sequence_0 = const()[name = tensor<string, []>("input_3_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_3_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("input_3_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_3_batch_first_cell_activation_0 = const()[name = tensor<string, []>("input_3_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_3_batch_first_activation_0 = const()[name = tensor<string, []>("input_3_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [1, 512]> input_3_batch_first_lstm_h0_reshaped_to_fp16 = const()[name = tensor<string, []>("input_3_batch_first_lstm_h0_reshaped_to_fp16"), val = tensor<fp16, [1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395008)))];
+            tensor<fp16, [1024, 640]> concat_5_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396096))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1051520))), name = tensor<string, []>("concat_5_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_6_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1052096))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1314304))), name = tensor<string, []>("concat_6_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_0_to_fp16 = const()[name = tensor<string, []>("add_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1314880)))];
+            tensor<fp16, [1024, 640]> concat_7_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1316992))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1972416))), name = tensor<string, []>("concat_7_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_8_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1972992))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2235200))), name = tensor<string, []>("concat_8_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_1_to_fp16 = const()[name = tensor<string, []>("add_1_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2235776)))];
+            tensor<fp16, [?, 1, 512]> input_3_batch_first_cast_fp16_0, tensor<fp16, [1, 512]> input_3_batch_first_cast_fp16_1, tensor<fp16, [1, 512]> input_3_batch_first_cast_fp16_2 = lstm(activation = input_3_batch_first_activation_0, bias = add_0_to_fp16, bias_back = add_1_to_fp16, cell_activation = input_3_batch_first_cell_activation_0, direction = input_3_batch_first_direction_0, initial_c = input_3_batch_first_lstm_h0_reshaped_to_fp16, initial_h = input_3_batch_first_lstm_h0_reshaped_to_fp16, output_sequence = input_3_batch_first_output_sequence_0, recurrent_activation = input_3_batch_first_recurrent_activation_0, weight_hh = concat_6_to_fp16_palettized, weight_hh_back = concat_8_to_fp16_palettized, weight_ih = concat_5_to_fp16_palettized, weight_ih_back = concat_7_to_fp16_palettized, x = x_7_cast_fp16)[name = tensor<string, []>("input_3_batch_first_cast_fp16")];
+            tensor<int32, [3]> transpose_17_perm_0 = const()[name = tensor<string, []>("transpose_17_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp16, [1024, 128]> dur_encoder_norms_0_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2237888))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2369024))), name = tensor<string, []>("dur_encoder_norms_0_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> dur_encoder_norms_0_fc_bias_to_fp16 = const()[name = tensor<string, []>("dur_encoder_norms_0_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2369600)))];
+            tensor<fp16, [1, 1024]> linear_1_cast_fp16 = linear(bias = dur_encoder_norms_0_fc_bias_to_fp16, weight = dur_encoder_norms_0_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_1_cast_fp16")];
+            tensor<int32, [3]> var_89 = const()[name = tensor<string, []>("op_89"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_3_cast_fp16 = reshape(shape = var_89, x = linear_1_cast_fp16)[name = tensor<string, []>("h_3_cast_fp16")];
+            tensor<int32, [2]> var_91_split_sizes_0 = const()[name = tensor<string, []>("op_91_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_91_axis_0 = const()[name = tensor<string, []>("op_91_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_91_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_91_cast_fp16_1 = split(axis = var_91_axis_0, split_sizes = var_91_split_sizes_0, x = h_3_cast_fp16)[name = tensor<string, []>("op_91_cast_fp16")];
+            tensor<int32, [3]> gamma_3_perm_0 = const()[name = tensor<string, []>("gamma_3_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [3]> beta_3_perm_0 = const()[name = tensor<string, []>("beta_3_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_21_axes_0 = const()[name = tensor<string, []>("x_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, []> var_20_to_fp16 = const()[name = tensor<string, []>("op_20_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, ?, 512]> transpose_17_cast_fp16 = transpose(perm = transpose_17_perm_0, x = input_3_batch_first_cast_fp16_0)[name = tensor<string, []>("transpose_45")];
+            tensor<fp16, [1, ?, 512]> x_21_cast_fp16 = layer_norm(axes = x_21_axes_0, epsilon = var_20_to_fp16, x = transpose_17_cast_fp16)[name = tensor<string, []>("x_21_cast_fp16")];
+            tensor<fp16, []> var_97_promoted_to_fp16 = const()[name = tensor<string, []>("op_97_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 1, 512]> gamma_3_cast_fp16 = transpose(perm = gamma_3_perm_0, x = var_91_cast_fp16_0)[name = tensor<string, []>("transpose_44")];
+            tensor<fp16, [1, 1, 512]> var_98_cast_fp16 = add(x = gamma_3_cast_fp16, y = var_97_promoted_to_fp16)[name = tensor<string, []>("op_98_cast_fp16")];
+            tensor<fp16, [1, ?, 512]> var_99_cast_fp16 = mul(x = var_98_cast_fp16, y = x_21_cast_fp16)[name = tensor<string, []>("op_99_cast_fp16")];
+            tensor<fp16, [1, 1, 512]> beta_3_cast_fp16 = transpose(perm = beta_3_perm_0, x = var_91_cast_fp16_1)[name = tensor<string, []>("transpose_43")];
+            tensor<fp16, [1, ?, 512]> x_23_cast_fp16 = add(x = var_99_cast_fp16, y = beta_3_cast_fp16)[name = tensor<string, []>("x_23_cast_fp16")];
+            tensor<bool, []> x_27_interleave_0 = const()[name = tensor<string, []>("x_27_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [3]> transpose_12_perm_0 = const()[name = tensor<string, []>("transpose_12_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<int32, [3]> transpose_13_perm_0 = const()[name = tensor<string, []>("transpose_13_perm_0"), val = tensor<int32, [3]>([1, 2, 0])];
+            tensor<fp16, [?, ?, ?]> transpose_13 = transpose(perm = transpose_13_perm_0, x = s_cast_fp16)[name = tensor<string, []>("transpose_41")];
+            tensor<fp16, [1, 512, ?]> transpose_12 = transpose(perm = transpose_12_perm_0, x = x_23_cast_fp16)[name = tensor<string, []>("transpose_42")];
+            tensor<fp16, [1, ?, ?]> x_27_cast_fp16 = concat(axis = var_31, interleave = x_27_interleave_0, values = (transpose_12, transpose_13))[name = tensor<string, []>("x_27_cast_fp16")];
+            tensor<int32, [3]> var_109_perm_0 = const()[name = tensor<string, []>("op_109_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<bool, [1, 1, ?]> var_109 = transpose(perm = var_109_perm_0, x = var_54)[name = tensor<string, []>("transpose_40")];
+            tensor<fp16, [1, ?, ?]> x_29_cast_fp16 = select(a = var_28_to_fp16, b = x_27_cast_fp16, cond = var_109)[name = tensor<string, []>("x_29_cast_fp16")];
+            tensor<int32, [3]> transpose_10_perm_0 = const()[name = tensor<string, []>("transpose_10_perm_0"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<string, []> input_9_batch_first_direction_0 = const()[name = tensor<string, []>("input_9_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> input_9_batch_first_output_sequence_0 = const()[name = tensor<string, []>("input_9_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_9_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("input_9_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_9_batch_first_cell_activation_0 = const()[name = tensor<string, []>("input_9_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_9_batch_first_activation_0 = const()[name = tensor<string, []>("input_9_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [1024, 640]> concat_15_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2371712))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3027136))), name = tensor<string, []>("concat_15_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_16_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3027712))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3289920))), name = tensor<string, []>("concat_16_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_2_to_fp16 = const()[name = tensor<string, []>("add_2_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3290496)))];
+            tensor<fp16, [1024, 640]> concat_17_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3292608))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3948032))), name = tensor<string, []>("concat_17_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_18_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3948608))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4210816))), name = tensor<string, []>("concat_18_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_3_to_fp16 = const()[name = tensor<string, []>("add_3_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4211392)))];
+            tensor<fp16, [?, 1, ?]> transpose_10_cast_fp16 = transpose(perm = transpose_10_perm_0, x = x_29_cast_fp16)[name = tensor<string, []>("transpose_39")];
+            tensor<fp16, [?, 1, 512]> input_9_batch_first_cast_fp16_0, tensor<fp16, [1, 512]> input_9_batch_first_cast_fp16_1, tensor<fp16, [1, 512]> input_9_batch_first_cast_fp16_2 = lstm(activation = input_9_batch_first_activation_0, bias = add_2_to_fp16, bias_back = add_3_to_fp16, cell_activation = input_9_batch_first_cell_activation_0, direction = input_9_batch_first_direction_0, initial_c = input_3_batch_first_lstm_h0_reshaped_to_fp16, initial_h = input_3_batch_first_lstm_h0_reshaped_to_fp16, output_sequence = input_9_batch_first_output_sequence_0, recurrent_activation = input_9_batch_first_recurrent_activation_0, weight_hh = concat_16_to_fp16_palettized, weight_hh_back = concat_18_to_fp16_palettized, weight_ih = concat_15_to_fp16_palettized, weight_ih_back = concat_17_to_fp16_palettized, x = transpose_10_cast_fp16)[name = tensor<string, []>("input_9_batch_first_cast_fp16")];
+            tensor<int32, [3]> transpose_18_perm_0 = const()[name = tensor<string, []>("transpose_18_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp16, [1024, 128]> dur_encoder_norms_1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4213504))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4344640))), name = tensor<string, []>("dur_encoder_norms_1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> dur_encoder_norms_1_fc_bias_to_fp16 = const()[name = tensor<string, []>("dur_encoder_norms_1_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4345216)))];
+            tensor<fp16, [1, 1024]> linear_2_cast_fp16 = linear(bias = dur_encoder_norms_1_fc_bias_to_fp16, weight = dur_encoder_norms_1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<int32, [3]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_7_cast_fp16 = reshape(shape = var_141, x = linear_2_cast_fp16)[name = tensor<string, []>("h_7_cast_fp16")];
+            tensor<int32, [2]> var_143_split_sizes_0 = const()[name = tensor<string, []>("op_143_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_143_axis_0 = const()[name = tensor<string, []>("op_143_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_143_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_143_cast_fp16_1 = split(axis = var_143_axis_0, split_sizes = var_143_split_sizes_0, x = h_7_cast_fp16)[name = tensor<string, []>("op_143_cast_fp16")];
+            tensor<int32, [3]> gamma_7_perm_0 = const()[name = tensor<string, []>("gamma_7_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [3]> beta_7_perm_0 = const()[name = tensor<string, []>("beta_7_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_39_axes_0 = const()[name = tensor<string, []>("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, ?, 512]> transpose_18_cast_fp16 = transpose(perm = transpose_18_perm_0, x = input_9_batch_first_cast_fp16_0)[name = tensor<string, []>("transpose_38")];
+            tensor<fp16, [1, ?, 512]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, epsilon = var_20_to_fp16, x = transpose_18_cast_fp16)[name = tensor<string, []>("x_39_cast_fp16")];
+            tensor<fp16, []> var_149_promoted_to_fp16 = const()[name = tensor<string, []>("op_149_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 1, 512]> gamma_7_cast_fp16 = transpose(perm = gamma_7_perm_0, x = var_143_cast_fp16_0)[name = tensor<string, []>("transpose_37")];
+            tensor<fp16, [1, 1, 512]> var_150_cast_fp16 = add(x = gamma_7_cast_fp16, y = var_149_promoted_to_fp16)[name = tensor<string, []>("op_150_cast_fp16")];
+            tensor<fp16, [1, ?, 512]> var_151_cast_fp16 = mul(x = var_150_cast_fp16, y = x_39_cast_fp16)[name = tensor<string, []>("op_151_cast_fp16")];
+            tensor<fp16, [1, 1, 512]> beta_7_cast_fp16 = transpose(perm = beta_7_perm_0, x = var_143_cast_fp16_1)[name = tensor<string, []>("transpose_36")];
+            tensor<fp16, [1, ?, 512]> x_41_cast_fp16 = add(x = var_151_cast_fp16, y = beta_7_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")];
+            tensor<bool, []> x_45_interleave_0 = const()[name = tensor<string, []>("x_45_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [3]> transpose_16_perm_0 = const()[name = tensor<string, []>("transpose_16_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<fp16, [1, 512, ?]> transpose_16 = transpose(perm = transpose_16_perm_0, x = x_41_cast_fp16)[name = tensor<string, []>("transpose_35")];
+            tensor<fp16, [1, ?, ?]> x_45_cast_fp16 = concat(axis = var_31, interleave = x_45_interleave_0, values = (transpose_16, transpose_13))[name = tensor<string, []>("x_45_cast_fp16")];
+            tensor<fp16, [1, ?, ?]> x_47_cast_fp16 = select(a = var_28_to_fp16, b = x_45_cast_fp16, cond = var_109)[name = tensor<string, []>("x_47_cast_fp16")];
+            tensor<int32, [3]> transpose_12_perm_0_1 = const()[name = tensor<string, []>("transpose_12_perm_0_1"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<string, []> input_15_batch_first_direction_0 = const()[name = tensor<string, []>("input_15_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> input_15_batch_first_output_sequence_0 = const()[name = tensor<string, []>("input_15_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_15_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("input_15_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_15_batch_first_cell_activation_0 = const()[name = tensor<string, []>("input_15_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_15_batch_first_activation_0 = const()[name = tensor<string, []>("input_15_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [1024, 640]> concat_25_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4347328))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5002752))), name = tensor<string, []>("concat_25_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_26_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5003328))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5265536))), name = tensor<string, []>("concat_26_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_4_to_fp16 = const()[name = tensor<string, []>("add_4_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5266112)))];
+            tensor<fp16, [1024, 640]> concat_27_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5268224))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5923648))), name = tensor<string, []>("concat_27_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_28_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5924224))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6186432))), name = tensor<string, []>("concat_28_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_5_to_fp16 = const()[name = tensor<string, []>("add_5_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6187008)))];
+            tensor<fp16, [?, 1, ?]> transpose_12_cast_fp16 = transpose(perm = transpose_12_perm_0_1, x = x_47_cast_fp16)[name = tensor<string, []>("transpose_34")];
+            tensor<fp16, [?, 1, 512]> input_15_batch_first_cast_fp16_0, tensor<fp16, [1, 512]> input_15_batch_first_cast_fp16_1, tensor<fp16, [1, 512]> input_15_batch_first_cast_fp16_2 = lstm(activation = input_15_batch_first_activation_0, bias = add_4_to_fp16, bias_back = add_5_to_fp16, cell_activation = input_15_batch_first_cell_activation_0, direction = input_15_batch_first_direction_0, initial_c = input_3_batch_first_lstm_h0_reshaped_to_fp16, initial_h = input_3_batch_first_lstm_h0_reshaped_to_fp16, output_sequence = input_15_batch_first_output_sequence_0, recurrent_activation = input_15_batch_first_recurrent_activation_0, weight_hh = concat_26_to_fp16_palettized, weight_hh_back = concat_28_to_fp16_palettized, weight_ih = concat_25_to_fp16_palettized, weight_ih_back = concat_27_to_fp16_palettized, x = transpose_12_cast_fp16)[name = tensor<string, []>("input_15_batch_first_cast_fp16")];
+            tensor<int32, [3]> transpose_19_perm_0 = const()[name = tensor<string, []>("transpose_19_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp16, [1024, 128]> dur_encoder_norms_2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6189120))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6320256))), name = tensor<string, []>("dur_encoder_norms_2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> dur_encoder_norms_2_fc_bias_to_fp16 = const()[name = tensor<string, []>("dur_encoder_norms_2_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6320832)))];
+            tensor<fp16, [1, 1024]> linear_3_cast_fp16 = linear(bias = dur_encoder_norms_2_fc_bias_to_fp16, weight = dur_encoder_norms_2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_3_cast_fp16")];
+            tensor<int32, [3]> var_193 = const()[name = tensor<string, []>("op_193"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_cast_fp16 = reshape(shape = var_193, x = linear_3_cast_fp16)[name = tensor<string, []>("h_cast_fp16")];
+            tensor<int32, [2]> var_195_split_sizes_0 = const()[name = tensor<string, []>("op_195_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_195_axis_0 = const()[name = tensor<string, []>("op_195_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_195_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_195_cast_fp16_1 = split(axis = var_195_axis_0, split_sizes = var_195_split_sizes_0, x = h_cast_fp16)[name = tensor<string, []>("op_195_cast_fp16")];
+            tensor<int32, [3]> gamma_11_perm_0 = const()[name = tensor<string, []>("gamma_11_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [3]> beta_11_perm_0 = const()[name = tensor<string, []>("beta_11_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_57_axes_0 = const()[name = tensor<string, []>("x_57_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, ?, 512]> transpose_19_cast_fp16 = transpose(perm = transpose_19_perm_0, x = input_15_batch_first_cast_fp16_0)[name = tensor<string, []>("transpose_33")];
+            tensor<fp16, [1, ?, 512]> x_57_cast_fp16 = layer_norm(axes = x_57_axes_0, epsilon = var_20_to_fp16, x = transpose_19_cast_fp16)[name = tensor<string, []>("x_57_cast_fp16")];
+            tensor<fp16, []> var_201_promoted_to_fp16 = const()[name = tensor<string, []>("op_201_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 1, 512]> gamma_11_cast_fp16 = transpose(perm = gamma_11_perm_0, x = var_195_cast_fp16_0)[name = tensor<string, []>("transpose_32")];
+            tensor<fp16, [1, 1, 512]> var_202_cast_fp16 = add(x = gamma_11_cast_fp16, y = var_201_promoted_to_fp16)[name = tensor<string, []>("op_202_cast_fp16")];
+            tensor<fp16, [1, ?, 512]> var_203_cast_fp16 = mul(x = var_202_cast_fp16, y = x_57_cast_fp16)[name = tensor<string, []>("op_203_cast_fp16")];
+            tensor<fp16, [1, 1, 512]> beta_11_cast_fp16 = transpose(perm = beta_11_perm_0, x = var_195_cast_fp16_1)[name = tensor<string, []>("transpose_31")];
+            tensor<fp16, [1, ?, 512]> x_59_cast_fp16 = add(x = var_203_cast_fp16, y = beta_11_cast_fp16)[name = tensor<string, []>("x_59_cast_fp16")];
+            tensor<bool, []> x_63_interleave_0 = const()[name = tensor<string, []>("x_63_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [3]> transpose_17_perm_0_1 = const()[name = tensor<string, []>("transpose_17_perm_0_1"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<fp16, [1, 512, ?]> transpose_17 = transpose(perm = transpose_17_perm_0_1, x = x_59_cast_fp16)[name = tensor<string, []>("transpose_30")];
+            tensor<fp16, [1, ?, ?]> x_63_cast_fp16 = concat(axis = var_31, interleave = x_63_interleave_0, values = (transpose_17, transpose_13))[name = tensor<string, []>("x_63_cast_fp16")];
+            tensor<fp16, [1, ?, ?]> x_65_cast_fp16 = select(a = var_28_to_fp16, b = x_63_cast_fp16, cond = var_109)[name = tensor<string, []>("x_65_cast_fp16")];
+            tensor<int32, [3]> input_19_perm_0 = const()[name = tensor<string, []>("input_19_perm_0"), val = tensor<int32, [3]>([0, -1, -2])];
+            tensor<int32, [3]> input_19_batch_first_transpose_perm_0 = const()[name = tensor<string, []>("input_19_batch_first_transpose_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<string, []> input_21_batch_first_direction_0 = const()[name = tensor<string, []>("input_21_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> input_21_batch_first_output_sequence_0 = const()[name = tensor<string, []>("input_21_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_21_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("input_21_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_21_batch_first_cell_activation_0 = const()[name = tensor<string, []>("input_21_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_21_batch_first_activation_0 = const()[name = tensor<string, []>("input_21_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [1024, 640]> concat_35_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6322944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6978368))), name = tensor<string, []>("concat_35_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_36_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6978944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7241152))), name = tensor<string, []>("concat_36_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_6_to_fp16 = const()[name = tensor<string, []>("add_6_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7241728)))];
+            tensor<fp16, [1024, 640]> concat_37_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7243840))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7899264))), name = tensor<string, []>("concat_37_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_38_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7899840))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8162048))), name = tensor<string, []>("concat_38_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_7_to_fp16 = const()[name = tensor<string, []>("add_7_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8162624)))];
+            tensor<fp16, [1, ?, ?]> d = transpose(perm = input_19_perm_0, x = x_65_cast_fp16)[name = tensor<string, []>("transpose_29")];
+            tensor<fp16, [?, 1, ?]> input_19_batch_first_transpose_cast_fp16 = transpose(perm = input_19_batch_first_transpose_perm_0, x = d)[name = tensor<string, []>("transpose_28")];
+            tensor<fp16, [?, 1, 512]> input_21_batch_first_cast_fp16_0, tensor<fp16, [1, 512]> input_21_batch_first_cast_fp16_1, tensor<fp16, [1, 512]> input_21_batch_first_cast_fp16_2 = lstm(activation = input_21_batch_first_activation_0, bias = add_6_to_fp16, bias_back = add_7_to_fp16, cell_activation = input_21_batch_first_cell_activation_0, direction = input_21_batch_first_direction_0, initial_c = input_3_batch_first_lstm_h0_reshaped_to_fp16, initial_h = input_3_batch_first_lstm_h0_reshaped_to_fp16, output_sequence = input_21_batch_first_output_sequence_0, recurrent_activation = input_21_batch_first_recurrent_activation_0, weight_hh = concat_36_to_fp16_palettized, weight_hh_back = concat_38_to_fp16_palettized, weight_ih = concat_35_to_fp16_palettized, weight_ih_back = concat_37_to_fp16_palettized, x = input_19_batch_first_transpose_cast_fp16)[name = tensor<string, []>("input_21_batch_first_cast_fp16")];
+            tensor<int32, [3]> input_21_perm_0 = const()[name = tensor<string, []>("input_21_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<fp16, [50, 512]> duration_proj_linear_layer_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8164736))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8190400))), name = tensor<string, []>("duration_proj_linear_layer_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([50, 512])];
+            tensor<fp16, [50]> duration_proj_linear_layer_bias_to_fp16 = const()[name = tensor<string, []>("duration_proj_linear_layer_bias_to_fp16"), val = tensor<fp16, [50]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8190976)))];
+            tensor<fp16, [1, ?, 512]> input_21_cast_fp16 = transpose(perm = input_21_perm_0, x = input_21_batch_first_cast_fp16_0)[name = tensor<string, []>("transpose_27")];
+            tensor<fp16, [1, ?, 50]> linear_4_cast_fp16 = linear(bias = duration_proj_linear_layer_bias_to_fp16, weight = duration_proj_linear_layer_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
+            tensor<fp16, [1, ?, 50]> var_248_cast_fp16 = sigmoid(x = linear_4_cast_fp16)[name = tensor<string, []>("op_248_cast_fp16")];
+            tensor<int32, [1]> var_253_axes_0 = const()[name = tensor<string, []>("op_253_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, []> var_253_keep_dims_0 = const()[name = tensor<string, []>("op_253_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, ?]> var_253_cast_fp16 = reduce_sum(axes = var_253_axes_0, keep_dims = var_253_keep_dims_0, x = var_248_cast_fp16)[name = tensor<string, []>("op_253_cast_fp16")];
+            tensor<fp16, [1, ?]> duration = real_div(x = var_253_cast_fp16, y = speed)[name = tensor<string, []>("op_254_cast_fp16")];
+            tensor<fp32, []> var_263 = const()[name = tensor<string, []>("op_263"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<int32, []> x_67_axis_0 = const()[name = tensor<string, []>("x_67_axis_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> x_67_batch_dims_0 = const()[name = tensor<string, []>("x_67_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> x_67_validate_indices_0 = const()[name = tensor<string, []>("x_67_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [178, 512]> text_encoder_embedding_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [91136]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8191168))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8282368))), name = tensor<string, []>("text_encoder_embedding_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([178, 512])];
+            tensor<string, []> input_ids_to_uint16_dtype_0 = const()[name = tensor<string, []>("input_ids_to_uint16_dtype_0"), val = tensor<string, []>("uint16")];
+            tensor<uint16, [1, ?]> input_ids_to_uint16 = cast(dtype = input_ids_to_uint16_dtype_0, x = input_ids)[name = tensor<string, []>("cast_0")];
+            tensor<fp16, [1, ?, 512]> x_67_cast_fp16_cast_uint16 = gather(axis = x_67_axis_0, batch_dims = x_67_batch_dims_0, indices = input_ids_to_uint16, validate_indices = x_67_validate_indices_0, x = text_encoder_embedding_weight_to_fp16_palettized)[name = tensor<string, []>("x_67_cast_fp16_cast_uint16")];
+            tensor<int32, [3]> x_69_perm_0 = const()[name = tensor<string, []>("x_69_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> m_unsq_axes_0 = const()[name = tensor<string, []>("m_unsq_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, ?]> m_unsq = expand_dims(axes = m_unsq_axes_0, x = m)[name = tensor<string, []>("m_unsq")];
+            tensor<fp16, []> var_265_to_fp16 = const()[name = tensor<string, []>("op_265_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 512, ?]> x_69_cast_fp16 = transpose(perm = x_69_perm_0, x = x_67_cast_fp16_cast_uint16)[name = tensor<string, []>("transpose_26")];
+            tensor<fp16, [1, 512, ?]> input_23_cast_fp16 = select(a = var_265_to_fp16, b = x_69_cast_fp16, cond = m_unsq)[name = tensor<string, []>("input_23_cast_fp16")];
+            tensor<string, []> x_71_pad_type_0 = const()[name = tensor<string, []>("x_71_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_71_pad_0 = const()[name = tensor<string, []>("x_71_pad_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [1]> x_71_strides_0 = const()[name = tensor<string, []>("x_71_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_71_dilations_0 = const()[name = tensor<string, []>("x_71_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_71_groups_0 = const()[name = tensor<string, []>("x_71_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 5]> weight_3_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8282944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9593728))), name = tensor<string, []>("weight_3_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 5])];
+            tensor<fp16, [512]> text_encoder_cnn_0_0_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_0_0_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9594304)))];
+            tensor<fp16, [1, 512, ?]> x_71_cast_fp16 = conv(bias = text_encoder_cnn_0_0_bias_to_fp16, dilations = x_71_dilations_0, groups = x_71_groups_0, pad = x_71_pad_0, pad_type = x_71_pad_type_0, strides = x_71_strides_0, weight = weight_3_to_fp16_palettized, x = input_23_cast_fp16)[name = tensor<string, []>("x_71_cast_fp16")];
+            tensor<int32, [3]> input_25_perm_0 = const()[name = tensor<string, []>("input_25_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_73_axes_0 = const()[name = tensor<string, []>("x_73_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [512]> text_encoder_cnn_0_1_gamma_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_0_1_gamma_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9595392)))];
+            tensor<fp16, [512]> text_encoder_cnn_0_1_beta_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_0_1_beta_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9596480)))];
+            tensor<fp16, []> var_261_to_fp16 = const()[name = tensor<string, []>("op_261_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, ?, 512]> input_25_cast_fp16 = transpose(perm = input_25_perm_0, x = x_71_cast_fp16)[name = tensor<string, []>("transpose_25")];
+            tensor<fp16, [1, ?, 512]> x_73_cast_fp16 = layer_norm(axes = x_73_axes_0, beta = text_encoder_cnn_0_1_beta_to_fp16, epsilon = var_261_to_fp16, gamma = text_encoder_cnn_0_1_gamma_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("x_73_cast_fp16")];
+            tensor<int32, [3]> input_27_perm_0 = const()[name = tensor<string, []>("input_27_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<fp16, [1, 512, ?]> input_27_cast_fp16 = transpose(perm = input_27_perm_0, x = x_73_cast_fp16)[name = tensor<string, []>("transpose_24")];
+            tensor<fp16, [1, 512, ?]> x_75_cast_fp16 = leaky_relu(alpha = var_263, x = input_27_cast_fp16)[name = tensor<string, []>("x_75_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_29_cast_fp16 = select(a = var_265_to_fp16, b = x_75_cast_fp16, cond = m_unsq)[name = tensor<string, []>("input_29_cast_fp16")];
+            tensor<string, []> x_77_pad_type_0 = const()[name = tensor<string, []>("x_77_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_77_pad_0 = const()[name = tensor<string, []>("x_77_pad_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [1]> x_77_strides_0 = const()[name = tensor<string, []>("x_77_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_77_dilations_0 = const()[name = tensor<string, []>("x_77_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_77_groups_0 = const()[name = tensor<string, []>("x_77_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 5]> weight_7_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9597568))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10908352))), name = tensor<string, []>("weight_7_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 5])];
+            tensor<fp16, [512]> text_encoder_cnn_1_0_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_1_0_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10908928)))];
+            tensor<fp16, [1, 512, ?]> x_77_cast_fp16 = conv(bias = text_encoder_cnn_1_0_bias_to_fp16, dilations = x_77_dilations_0, groups = x_77_groups_0, pad = x_77_pad_0, pad_type = x_77_pad_type_0, strides = x_77_strides_0, weight = weight_7_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("x_77_cast_fp16")];
+            tensor<int32, [3]> input_31_perm_0 = const()[name = tensor<string, []>("input_31_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_79_axes_0 = const()[name = tensor<string, []>("x_79_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [512]> text_encoder_cnn_1_1_gamma_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_1_1_gamma_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10910016)))];
+            tensor<fp16, [512]> text_encoder_cnn_1_1_beta_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_1_1_beta_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10911104)))];
+            tensor<fp16, [1, ?, 512]> input_31_cast_fp16 = transpose(perm = input_31_perm_0, x = x_77_cast_fp16)[name = tensor<string, []>("transpose_23")];
+            tensor<fp16, [1, ?, 512]> x_79_cast_fp16 = layer_norm(axes = x_79_axes_0, beta = text_encoder_cnn_1_1_beta_to_fp16, epsilon = var_261_to_fp16, gamma = text_encoder_cnn_1_1_gamma_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("x_79_cast_fp16")];
+            tensor<int32, [3]> input_33_perm_0 = const()[name = tensor<string, []>("input_33_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<fp16, [1, 512, ?]> input_33_cast_fp16 = transpose(perm = input_33_perm_0, x = x_79_cast_fp16)[name = tensor<string, []>("transpose_22")];
+            tensor<fp16, [1, 512, ?]> x_81_cast_fp16 = leaky_relu(alpha = var_263, x = input_33_cast_fp16)[name = tensor<string, []>("x_81_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_35_cast_fp16 = select(a = var_265_to_fp16, b = x_81_cast_fp16, cond = m_unsq)[name = tensor<string, []>("input_35_cast_fp16")];
+            tensor<string, []> x_83_pad_type_0 = const()[name = tensor<string, []>("x_83_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_83_pad_0 = const()[name = tensor<string, []>("x_83_pad_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [1]> x_83_strides_0 = const()[name = tensor<string, []>("x_83_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_83_dilations_0 = const()[name = tensor<string, []>("x_83_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_83_groups_0 = const()[name = tensor<string, []>("x_83_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 5]> weight_11_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1310720]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10912192))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12222976))), name = tensor<string, []>("weight_11_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 5])];
+            tensor<fp16, [512]> text_encoder_cnn_2_0_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_2_0_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12223552)))];
+            tensor<fp16, [1, 512, ?]> x_83_cast_fp16 = conv(bias = text_encoder_cnn_2_0_bias_to_fp16, dilations = x_83_dilations_0, groups = x_83_groups_0, pad = x_83_pad_0, pad_type = x_83_pad_type_0, strides = x_83_strides_0, weight = weight_11_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor<string, []>("x_83_cast_fp16")];
+            tensor<int32, [3]> input_37_perm_0 = const()[name = tensor<string, []>("input_37_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<int32, [1]> x_85_axes_0 = const()[name = tensor<string, []>("x_85_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [512]> text_encoder_cnn_2_1_gamma_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_2_1_gamma_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12224640)))];
+            tensor<fp16, [512]> text_encoder_cnn_2_1_beta_to_fp16 = const()[name = tensor<string, []>("text_encoder_cnn_2_1_beta_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12225728)))];
+            tensor<fp16, [1, ?, 512]> input_37_cast_fp16 = transpose(perm = input_37_perm_0, x = x_83_cast_fp16)[name = tensor<string, []>("transpose_21")];
+            tensor<fp16, [1, ?, 512]> x_85_cast_fp16 = layer_norm(axes = x_85_axes_0, beta = text_encoder_cnn_2_1_beta_to_fp16, epsilon = var_261_to_fp16, gamma = text_encoder_cnn_2_1_gamma_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("x_85_cast_fp16")];
+            tensor<int32, [3]> input_39_perm_0 = const()[name = tensor<string, []>("input_39_perm_0"), val = tensor<int32, [3]>([0, -1, 1])];
+            tensor<fp16, [1, 512, ?]> input_39_cast_fp16 = transpose(perm = input_39_perm_0, x = x_85_cast_fp16)[name = tensor<string, []>("transpose_20")];
+            tensor<fp16, [1, 512, ?]> x_87_cast_fp16 = leaky_relu(alpha = var_263, x = input_39_cast_fp16)[name = tensor<string, []>("x_87_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> x_89_cast_fp16 = select(a = var_265_to_fp16, b = x_87_cast_fp16, cond = m_unsq)[name = tensor<string, []>("x_89_cast_fp16")];
+            tensor<int32, [3]> transpose_14_perm_0 = const()[name = tensor<string, []>("transpose_14_perm_0"), val = tensor<int32, [3]>([2, 0, 1])];
+            tensor<string, []> x_91_batch_first_direction_0 = const()[name = tensor<string, []>("x_91_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> x_91_batch_first_output_sequence_0 = const()[name = tensor<string, []>("x_91_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> x_91_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("x_91_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> x_91_batch_first_cell_activation_0 = const()[name = tensor<string, []>("x_91_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> x_91_batch_first_activation_0 = const()[name = tensor<string, []>("x_91_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [1024, 512]> concat_45_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12226816))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12751168))), name = tensor<string, []>("concat_45_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 512])];
+            tensor<fp16, [1024, 256]> concat_46_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12751744))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13013952))), name = tensor<string, []>("concat_46_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_8_to_fp16 = const()[name = tensor<string, []>("add_8_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13014528)))];
+            tensor<fp16, [1024, 512]> concat_47_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13016640))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13540992))), name = tensor<string, []>("concat_47_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 512])];
+            tensor<fp16, [1024, 256]> concat_48_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13541568))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13803776))), name = tensor<string, []>("concat_48_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_9_to_fp16 = const()[name = tensor<string, []>("add_9_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13804352)))];
+            tensor<fp16, [?, 1, 512]> transpose_14_cast_fp16 = transpose(perm = transpose_14_perm_0, x = x_89_cast_fp16)[name = tensor<string, []>("transpose_19")];
+            tensor<fp16, [?, 1, 512]> x_91_batch_first_cast_fp16_0, tensor<fp16, [1, 512]> x_91_batch_first_cast_fp16_1, tensor<fp16, [1, 512]> x_91_batch_first_cast_fp16_2 = lstm(activation = x_91_batch_first_activation_0, bias = add_8_to_fp16, bias_back = add_9_to_fp16, cell_activation = x_91_batch_first_cell_activation_0, direction = x_91_batch_first_direction_0, initial_c = input_3_batch_first_lstm_h0_reshaped_to_fp16, initial_h = input_3_batch_first_lstm_h0_reshaped_to_fp16, output_sequence = x_91_batch_first_output_sequence_0, recurrent_activation = x_91_batch_first_recurrent_activation_0, weight_hh = concat_46_to_fp16_palettized, weight_hh_back = concat_48_to_fp16_palettized, weight_ih = concat_45_to_fp16_palettized, weight_ih_back = concat_47_to_fp16_palettized, x = transpose_14_cast_fp16)[name = tensor<string, []>("x_91_batch_first_cast_fp16")];
+            tensor<int32, [3]> transpose_15_perm_0 = const()[name = tensor<string, []>("transpose_15_perm_0"), val = tensor<int32, [3]>([1, 2, 0])];
+            tensor<fp16, [1, 512, ?]> transpose_15_cast_fp16 = transpose(perm = transpose_15_perm_0, x = x_91_batch_first_cast_fp16_0)[name = tensor<string, []>("transpose_18")];
+            tensor<fp16, [1, 512, ?]> t_en = select(a = var_265_to_fp16, b = transpose_15_cast_fp16, cond = m_unsq)[name = tensor<string, []>("t_en_cast_fp16")];
+        } -> (duration, d, t_en);
+}

ANE/ANE-zh/KokoroPostAlbert.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bca73c9f402ccd18052fe8cc418741caa196c3caceeb3637187cdc5d67d8daf7
+size 13806464

ANE/ANE-zh/KokoroPostAlbert.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5669010b4feed54b3e9431ae12983264a3e09be548ce6d08c3e02bbb5c1f6a0
+size 49183

ANE/ANE-zh/KokoroPostAlbert.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bca73c9f402ccd18052fe8cc418741caa196c3caceeb3637187cdc5d67d8daf7
+size 13806464

ANE/ANE-zh/KokoroPostAlbert.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "07F36BCD-E5BB-4939-9D39-45F59CBD8058": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "DB17D18A-94F8-4785-9B53-34E1386398F3": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "DB17D18A-94F8-4785-9B53-34E1386398F3"
+}

ANE/ANE-zh/KokoroProsody.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c90a008bfe021f8c24f39d26706a894cffb3cc6d5dca1ad131f54547f0554a1
+size 243

ANE/ANE-zh/KokoroProsody.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f1932bc07f52171e0fff898ae0e9c289a8c73aaf0a2ac06b2352b8ef58afc61
+size 422

ANE/ANE-zh/KokoroProsody.mlmodelc/metadata.json

@@ -0,0 +1,98 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Mixed (Float16, Palettized (8 bits))",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "F0",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "N",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.mul" : 18,
+      "Ios17.linear" : 12,
+      "Ios17.transpose" : 2,
+      "Ios16.constexprLutToDense" : 30,
+      "Ios17.conv" : 16,
+      "Ios17.leakyRelu" : 12,
+      "Ios17.add" : 30,
+      "Ios17.convTranspose" : 2,
+      "Ios17.lstm" : 1,
+      "Ios17.sliceByIndex" : 2,
+      "UpsampleNearestNeighbor" : 2,
+      "Ios17.expandDims" : 2,
+      "Ios17.instanceNorm" : 11,
+      "Ios17.reshape" : 12,
+      "Split" : 12,
+      "Ios17.squeeze" : 4
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-05-03",
+      "com.github.apple.coremltools.source" : "torch==2.11.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float16",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 640 × 1...2000",
+        "shapeRange" : "[[1, 1], [640, 640], [1, 2000]]",
+        "formattedType" : "MultiArray (Float16 1 × 640 × 133)",
+        "type" : "MultiArray",
+        "shape" : "[1, 640, 133]",
+        "name" : "en",
+        "shortDescription" : ""
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 128]",
+        "name" : "style_s",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "KokoroProsody",
+    "method" : "predict"
+  }
+]

ANE/ANE-zh/KokoroProsody.mlmodelc/model.mil

@@ -0,0 +1,394 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
+{
+    func main<ios17>(tensor<fp16, [1, 640, ?]> en, tensor<fp16, [1, 128]> style_s) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"en", [1, 640, 133]}}), ("RangeDims", {{"en", [[1, 1], [640, 640], [1, 2000]]}})))] {
+            tensor<int32, [3]> transpose_0_perm_0 = const()[name = tensor<string, []>("transpose_0_perm_0"), val = tensor<int32, [3]>([-1, 0, -2])];
+            tensor<string, []> x_batch_first_direction_0 = const()[name = tensor<string, []>("x_batch_first_direction_0"), val = tensor<string, []>("bidirectional")];
+            tensor<bool, []> x_batch_first_output_sequence_0 = const()[name = tensor<string, []>("x_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> x_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("x_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> x_batch_first_cell_activation_0 = const()[name = tensor<string, []>("x_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> x_batch_first_activation_0 = const()[name = tensor<string, []>("x_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [1, 512]> x_batch_first_lstm_h0_reshaped_to_fp16 = const()[name = tensor<string, []>("x_batch_first_lstm_h0_reshaped_to_fp16"), val = tensor<fp16, [1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1024, 640]> concat_4_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1152))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(656576))), name = tensor<string, []>("concat_4_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_5_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657152))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919360))), name = tensor<string, []>("concat_5_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_0_to_fp16 = const()[name = tensor<string, []>("add_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919936)))];
+            tensor<fp16, [1024, 640]> concat_6_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [655360]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922048))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1577472))), name = tensor<string, []>("concat_6_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 640])];
+            tensor<fp16, [1024, 256]> concat_7_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1578048))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1840256))), name = tensor<string, []>("concat_7_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 256])];
+            tensor<fp16, [1024]> add_1_to_fp16 = const()[name = tensor<string, []>("add_1_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1840832)))];
+            tensor<fp16, [?, 1, 640]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = en)[name = tensor<string, []>("transpose_2")];
+            tensor<fp16, [?, 1, 512]> x_batch_first_cast_fp16_0, tensor<fp16, [1, 512]> x_batch_first_cast_fp16_1, tensor<fp16, [1, 512]> x_batch_first_cast_fp16_2 = lstm(activation = x_batch_first_activation_0, bias = add_0_to_fp16, bias_back = add_1_to_fp16, cell_activation = x_batch_first_cell_activation_0, direction = x_batch_first_direction_0, initial_c = x_batch_first_lstm_h0_reshaped_to_fp16, initial_h = x_batch_first_lstm_h0_reshaped_to_fp16, output_sequence = x_batch_first_output_sequence_0, recurrent_activation = x_batch_first_recurrent_activation_0, weight_hh = concat_5_to_fp16_palettized, weight_hh_back = concat_7_to_fp16_palettized, weight_ih = concat_4_to_fp16_palettized, weight_ih_back = concat_6_to_fp16_palettized, x = transpose_0_cast_fp16)[name = tensor<string, []>("x_batch_first_cast_fp16")];
+            tensor<fp32, []> var_53 = const()[name = tensor<string, []>("op_53"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp16, [1024, 128]> F0_0_norm1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1842944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1974080))), name = tensor<string, []>("F0_0_norm1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> F0_0_norm1_fc_bias_to_fp16 = const()[name = tensor<string, []>("F0_0_norm1_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1974656)))];
+            tensor<fp16, [1, 1024]> linear_0_cast_fp16 = linear(bias = F0_0_norm1_fc_bias_to_fp16, weight = F0_0_norm1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<int32, [3]> var_79 = const()[name = tensor<string, []>("op_79"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_3_cast_fp16 = reshape(shape = var_79, x = linear_0_cast_fp16)[name = tensor<string, []>("h_3_cast_fp16")];
+            tensor<int32, [2]> var_81_split_sizes_0 = const()[name = tensor<string, []>("op_81_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_81_axis_0 = const()[name = tensor<string, []>("op_81_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_81_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_81_cast_fp16_1 = split(axis = var_81_axis_0, split_sizes = var_81_split_sizes_0, x = h_3_cast_fp16)[name = tensor<string, []>("op_81_cast_fp16")];
+            tensor<fp16, []> var_83_promoted_to_fp16 = const()[name = tensor<string, []>("op_83_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 512, 1]> var_84_cast_fp16 = add(x = var_81_cast_fp16_0, y = var_83_promoted_to_fp16)[name = tensor<string, []>("op_84_cast_fp16")];
+            tensor<fp16, [512]> F0_0_norm1_norm_weight_to_fp16 = const()[name = tensor<string, []>("F0_0_norm1_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1976768)))];
+            tensor<fp16, [512]> F0_0_norm1_norm_bias_to_fp16 = const()[name = tensor<string, []>("F0_0_norm1_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1977856)))];
+            tensor<fp16, []> var_56_to_fp16 = const()[name = tensor<string, []>("op_56_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<int32, [3]> transpose_0_perm_0_1 = const()[name = tensor<string, []>("transpose_0_perm_0_1"), val = tensor<int32, [3]>([1, 2, 0])];
+            tensor<fp16, [1, 512, ?]> transpose_0 = transpose(perm = transpose_0_perm_0_1, x = x_batch_first_cast_fp16_0)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 512, ?]> var_87_cast_fp16 = instance_norm(beta = F0_0_norm1_norm_bias_to_fp16, epsilon = var_56_to_fp16, gamma = F0_0_norm1_norm_weight_to_fp16, x = transpose_0)[name = tensor<string, []>("op_87_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_88_cast_fp16 = mul(x = var_84_cast_fp16, y = var_87_cast_fp16)[name = tensor<string, []>("op_88_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_5_cast_fp16 = add(x = var_88_cast_fp16, y = var_81_cast_fp16_1)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_7_cast_fp16 = leaky_relu(alpha = var_53, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
+            tensor<string, []> input_9_pad_type_0 = const()[name = tensor<string, []>("input_9_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_9_pad_0 = const()[name = tensor<string, []>("input_9_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_9_strides_0 = const()[name = tensor<string, []>("input_9_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_9_dilations_0 = const()[name = tensor<string, []>("input_9_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_9_groups_0 = const()[name = tensor<string, []>("input_9_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 3]> weight_3_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1978944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2765440))), name = tensor<string, []>("weight_3_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 3])];
+            tensor<fp16, [512]> F0_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("F0_0_conv1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2766016)))];
+            tensor<fp16, [1, 512, ?]> input_9_cast_fp16 = conv(bias = F0_0_conv1_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = weight_3_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
+            tensor<fp16, [1024, 128]> F0_0_norm2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2767104))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2898240))), name = tensor<string, []>("F0_0_norm2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> F0_0_norm2_fc_bias_to_fp16 = const()[name = tensor<string, []>("F0_0_norm2_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2898816)))];
+            tensor<fp16, [1, 1024]> linear_1_cast_fp16 = linear(bias = F0_0_norm2_fc_bias_to_fp16, weight = F0_0_norm2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_1_cast_fp16")];
+            tensor<int32, [3]> var_105 = const()[name = tensor<string, []>("op_105"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_7_cast_fp16 = reshape(shape = var_105, x = linear_1_cast_fp16)[name = tensor<string, []>("h_7_cast_fp16")];
+            tensor<int32, [2]> var_107_split_sizes_0 = const()[name = tensor<string, []>("op_107_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_107_axis_0 = const()[name = tensor<string, []>("op_107_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_107_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_107_cast_fp16_1 = split(axis = var_107_axis_0, split_sizes = var_107_split_sizes_0, x = h_7_cast_fp16)[name = tensor<string, []>("op_107_cast_fp16")];
+            tensor<fp16, []> var_109_promoted_to_fp16 = const()[name = tensor<string, []>("op_109_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 512, 1]> var_110_cast_fp16 = add(x = var_107_cast_fp16_0, y = var_109_promoted_to_fp16)[name = tensor<string, []>("op_110_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_113_cast_fp16 = instance_norm(beta = F0_0_norm1_norm_bias_to_fp16, epsilon = var_56_to_fp16, gamma = F0_0_norm1_norm_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("op_113_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_114_cast_fp16 = mul(x = var_110_cast_fp16, y = var_113_cast_fp16)[name = tensor<string, []>("op_114_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_11_cast_fp16 = add(x = var_114_cast_fp16, y = var_107_cast_fp16_1)[name = tensor<string, []>("input_11_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_13_cast_fp16 = leaky_relu(alpha = var_53, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
+            tensor<string, []> out_1_pad_type_0 = const()[name = tensor<string, []>("out_1_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> out_1_pad_0 = const()[name = tensor<string, []>("out_1_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> out_1_strides_0 = const()[name = tensor<string, []>("out_1_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> out_1_dilations_0 = const()[name = tensor<string, []>("out_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> out_1_groups_0 = const()[name = tensor<string, []>("out_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 3]> weight_7_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2900928))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3687424))), name = tensor<string, []>("weight_7_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 3])];
+            tensor<fp16, [512]> F0_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("F0_0_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3688000)))];
+            tensor<fp16, [1, 512, ?]> out_1_cast_fp16 = conv(bias = F0_0_conv2_bias_to_fp16, dilations = out_1_dilations_0, groups = out_1_groups_0, pad = out_1_pad_0, pad_type = out_1_pad_type_0, strides = out_1_strides_0, weight = weight_7_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_124_cast_fp16 = add(x = out_1_cast_fp16, y = transpose_0)[name = tensor<string, []>("op_124_cast_fp16")];
+            tensor<fp16, []> var_125_to_fp16 = const()[name = tensor<string, []>("op_125_to_fp16"), val = tensor<fp16, []>(0x1.6ap-1)];
+            tensor<fp16, [1, 512, ?]> input_15_cast_fp16 = mul(x = var_124_cast_fp16, y = var_125_to_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
+            tensor<fp32, []> var_130 = const()[name = tensor<string, []>("op_130"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp16, [1024, 128]> F0_1_norm1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3689088))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3820224))), name = tensor<string, []>("F0_1_norm1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> F0_1_norm1_fc_bias_to_fp16 = const()[name = tensor<string, []>("F0_1_norm1_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3820800)))];
+            tensor<fp16, [1, 1024]> linear_2_cast_fp16 = linear(bias = F0_1_norm1_fc_bias_to_fp16, weight = F0_1_norm1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<int32, [3]> var_166 = const()[name = tensor<string, []>("op_166"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_11_cast_fp16 = reshape(shape = var_166, x = linear_2_cast_fp16)[name = tensor<string, []>("h_11_cast_fp16")];
+            tensor<int32, [2]> var_168_split_sizes_0 = const()[name = tensor<string, []>("op_168_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_168_axis_0 = const()[name = tensor<string, []>("op_168_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_168_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_168_cast_fp16_1 = split(axis = var_168_axis_0, split_sizes = var_168_split_sizes_0, x = h_11_cast_fp16)[name = tensor<string, []>("op_168_cast_fp16")];
+            tensor<fp16, []> var_170_promoted_to_fp16 = const()[name = tensor<string, []>("op_170_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 512, 1]> var_171_cast_fp16 = add(x = var_168_cast_fp16_0, y = var_170_promoted_to_fp16)[name = tensor<string, []>("op_171_cast_fp16")];
+            tensor<fp16, []> var_134_to_fp16 = const()[name = tensor<string, []>("op_134_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 512, ?]> var_174_cast_fp16 = instance_norm(beta = F0_0_norm1_norm_bias_to_fp16, epsilon = var_134_to_fp16, gamma = F0_0_norm1_norm_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("op_174_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_175_cast_fp16 = mul(x = var_171_cast_fp16, y = var_174_cast_fp16)[name = tensor<string, []>("op_175_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_17_cast_fp16 = add(x = var_175_cast_fp16, y = var_168_cast_fp16_1)[name = tensor<string, []>("input_17_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_19_cast_fp16 = leaky_relu(alpha = var_130, x = input_17_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
+            tensor<string, []> conv_transpose_0_pad_type_0 = const()[name = tensor<string, []>("conv_transpose_0_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> conv_transpose_0_pad_0 = const()[name = tensor<string, []>("conv_transpose_0_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_transpose_0_strides_0 = const()[name = tensor<string, []>("conv_transpose_0_strides_0"), val = tensor<int32, [1]>([2])];
+            tensor<int32, []> conv_transpose_0_groups_0 = const()[name = tensor<string, []>("conv_transpose_0_groups_0"), val = tensor<int32, []>(512)];
+            tensor<int32, [1]> conv_transpose_0_dilations_0 = const()[name = tensor<string, []>("conv_transpose_0_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [512, 1, 3]> var_178_to_fp16 = const()[name = tensor<string, []>("op_178_to_fp16"), val = tensor<fp16, [512, 1, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3822912)))];
+            tensor<fp16, [512]> F0_1_pool_bias_to_fp16 = const()[name = tensor<string, []>("F0_1_pool_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3826048)))];
+            tensor<fp16, [1, 512, ?]> conv_transpose_0_cast_fp16 = conv_transpose(bias = F0_1_pool_bias_to_fp16, dilations = conv_transpose_0_dilations_0, groups = conv_transpose_0_groups_0, pad = conv_transpose_0_pad_0, pad_type = conv_transpose_0_pad_type_0, strides = conv_transpose_0_strides_0, weight = var_178_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("conv_transpose_0_cast_fp16")];
+            tensor<int32, [3]> input_21_begin_0 = const()[name = tensor<string, []>("input_21_begin_0"), val = tensor<int32, [3]>([0, 0, 1])];
+            tensor<int32, [3]> input_21_end_0 = const()[name = tensor<string, []>("input_21_end_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<bool, [3]> input_21_begin_mask_0 = const()[name = tensor<string, []>("input_21_begin_mask_0"), val = tensor<bool, [3]>([true, true, false])];
+            tensor<bool, [3]> input_21_end_mask_0 = const()[name = tensor<string, []>("input_21_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
+            tensor<fp16, [1, 512, ?]> input_21_cast_fp16 = slice_by_index(begin = input_21_begin_0, begin_mask = input_21_begin_mask_0, end = input_21_end_0, end_mask = input_21_end_mask_0, x = conv_transpose_0_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
+            tensor<string, []> input_23_pad_type_0 = const()[name = tensor<string, []>("input_23_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_23_pad_0 = const()[name = tensor<string, []>("input_23_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_23_strides_0 = const()[name = tensor<string, []>("input_23_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_23_dilations_0 = const()[name = tensor<string, []>("input_23_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_23_groups_0 = const()[name = tensor<string, []>("input_23_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 512, 3]> weight_11_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [393216]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3827136))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4220416))), name = tensor<string, []>("weight_11_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 512, 3])];
+            tensor<fp16, [256]> F0_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("F0_1_conv1_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4220992)))];
+            tensor<fp16, [1, 256, ?]> input_23_cast_fp16 = conv(bias = F0_1_conv1_bias_to_fp16, dilations = input_23_dilations_0, groups = input_23_groups_0, pad = input_23_pad_0, pad_type = input_23_pad_type_0, strides = input_23_strides_0, weight = weight_11_to_fp16_palettized, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")];
+            tensor<fp16, [512, 128]> F0_1_norm2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4221568))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4287168))), name = tensor<string, []>("F0_1_norm2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp16, [512]> F0_1_norm2_fc_bias_to_fp16 = const()[name = tensor<string, []>("F0_1_norm2_fc_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4287744)))];
+            tensor<fp16, [1, 512]> linear_3_cast_fp16 = linear(bias = F0_1_norm2_fc_bias_to_fp16, weight = F0_1_norm2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_3_cast_fp16")];
+            tensor<int32, [3]> var_199 = const()[name = tensor<string, []>("op_199"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp16, [1, 512, 1]> h_15_cast_fp16 = reshape(shape = var_199, x = linear_3_cast_fp16)[name = tensor<string, []>("h_15_cast_fp16")];
+            tensor<int32, [2]> var_201_split_sizes_0 = const()[name = tensor<string, []>("op_201_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_201_axis_0 = const()[name = tensor<string, []>("op_201_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> var_201_cast_fp16_0, tensor<fp16, [1, 256, 1]> var_201_cast_fp16_1 = split(axis = var_201_axis_0, split_sizes = var_201_split_sizes_0, x = h_15_cast_fp16)[name = tensor<string, []>("op_201_cast_fp16")];
+            tensor<fp16, []> var_203_promoted_to_fp16 = const()[name = tensor<string, []>("op_203_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 256, 1]> var_204_cast_fp16 = add(x = var_201_cast_fp16_0, y = var_203_promoted_to_fp16)[name = tensor<string, []>("op_204_cast_fp16")];
+            tensor<fp16, [256]> F0_1_norm2_norm_weight_to_fp16 = const()[name = tensor<string, []>("F0_1_norm2_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4288832)))];
+            tensor<fp16, [256]> F0_1_norm2_norm_bias_to_fp16 = const()[name = tensor<string, []>("F0_1_norm2_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4289408)))];
+            tensor<fp16, [1, 256, ?]> var_207_cast_fp16 = instance_norm(beta = F0_1_norm2_norm_bias_to_fp16, epsilon = var_134_to_fp16, gamma = F0_1_norm2_norm_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("op_207_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_208_cast_fp16 = mul(x = var_204_cast_fp16, y = var_207_cast_fp16)[name = tensor<string, []>("op_208_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_25_cast_fp16 = add(x = var_208_cast_fp16, y = var_201_cast_fp16_1)[name = tensor<string, []>("input_25_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_27_cast_fp16 = leaky_relu(alpha = var_130, x = input_25_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
+            tensor<string, []> out_3_pad_type_0 = const()[name = tensor<string, []>("out_3_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> out_3_pad_0 = const()[name = tensor<string, []>("out_3_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> out_3_strides_0 = const()[name = tensor<string, []>("out_3_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> out_3_dilations_0 = const()[name = tensor<string, []>("out_3_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> out_3_groups_0 = const()[name = tensor<string, []>("out_3_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 256, 3]> weight_15_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [196608]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4289984))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4486656))), name = tensor<string, []>("weight_15_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 256, 3])];
+            tensor<fp16, [256]> F0_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("F0_1_conv2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4487232)))];
+            tensor<fp16, [1, 256, ?]> out_3_cast_fp16 = conv(bias = F0_1_conv2_bias_to_fp16, dilations = out_3_dilations_0, groups = out_3_groups_0, pad = out_3_pad_0, pad_type = out_3_pad_type_0, strides = out_3_strides_0, weight = weight_15_to_fp16_palettized, x = input_27_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp16, [1, 512, ?, 1]> expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_15_cast_fp16)[name = tensor<string, []>("expand_dims_0_cast_fp16")];
+            tensor<int32, []> upsample_nearest_neighbor_0_scale_factor_height_0 = const()[name = tensor<string, []>("upsample_nearest_neighbor_0_scale_factor_height_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> upsample_nearest_neighbor_0_scale_factor_width_0 = const()[name = tensor<string, []>("upsample_nearest_neighbor_0_scale_factor_width_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, ?, 1]> upsample_nearest_neighbor_0_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = upsample_nearest_neighbor_0_scale_factor_height_0, scale_factor_width = upsample_nearest_neighbor_0_scale_factor_width_0, x = expand_dims_0_cast_fp16)[name = tensor<string, []>("upsample_nearest_neighbor_0_cast_fp16")];
+            tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp16, [1, 512, ?]> input_29_cast_fp16 = squeeze(axes = input_29_axes_0, x = upsample_nearest_neighbor_0_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
+            tensor<string, []> var_225_pad_type_0 = const()[name = tensor<string, []>("op_225_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_225_strides_0 = const()[name = tensor<string, []>("op_225_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_225_pad_0 = const()[name = tensor<string, []>("op_225_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_225_dilations_0 = const()[name = tensor<string, []>("op_225_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_225_groups_0 = const()[name = tensor<string, []>("op_225_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 512, 1]> weight_17_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4487808))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4618944))), name = tensor<string, []>("weight_17_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 512, 1])];
+            tensor<fp16, [1, 256, ?]> var_225_cast_fp16 = conv(dilations = var_225_dilations_0, groups = var_225_groups_0, pad = var_225_pad_0, pad_type = var_225_pad_type_0, strides = var_225_strides_0, weight = weight_17_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("op_225_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_226_cast_fp16 = add(x = out_3_cast_fp16, y = var_225_cast_fp16)[name = tensor<string, []>("op_226_cast_fp16")];
+            tensor<fp16, []> var_227_to_fp16 = const()[name = tensor<string, []>("op_227_to_fp16"), val = tensor<fp16, []>(0x1.6ap-1)];
+            tensor<fp16, [1, 256, ?]> input_31_cast_fp16 = mul(x = var_226_cast_fp16, y = var_227_to_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
+            tensor<fp32, []> var_230 = const()[name = tensor<string, []>("op_230"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp16, [512, 128]> F0_2_norm1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4619520))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4685120))), name = tensor<string, []>("F0_2_norm1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp16, [512]> F0_2_norm1_fc_bias_to_fp16 = const()[name = tensor<string, []>("F0_2_norm1_fc_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4685696)))];
+            tensor<fp16, [1, 512]> linear_4_cast_fp16 = linear(bias = F0_2_norm1_fc_bias_to_fp16, weight = F0_2_norm1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_4_cast_fp16")];
+            tensor<int32, [3]> var_256 = const()[name = tensor<string, []>("op_256"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp16, [1, 512, 1]> h_19_cast_fp16 = reshape(shape = var_256, x = linear_4_cast_fp16)[name = tensor<string, []>("h_19_cast_fp16")];
+            tensor<int32, [2]> var_258_split_sizes_0 = const()[name = tensor<string, []>("op_258_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_258_axis_0 = const()[name = tensor<string, []>("op_258_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> var_258_cast_fp16_0, tensor<fp16, [1, 256, 1]> var_258_cast_fp16_1 = split(axis = var_258_axis_0, split_sizes = var_258_split_sizes_0, x = h_19_cast_fp16)[name = tensor<string, []>("op_258_cast_fp16")];
+            tensor<fp16, []> var_260_promoted_to_fp16 = const()[name = tensor<string, []>("op_260_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 256, 1]> var_261_cast_fp16 = add(x = var_258_cast_fp16_0, y = var_260_promoted_to_fp16)[name = tensor<string, []>("op_261_cast_fp16")];
+            tensor<fp16, []> var_233_to_fp16 = const()[name = tensor<string, []>("op_233_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 256, ?]> var_264_cast_fp16 = instance_norm(beta = F0_1_norm2_norm_bias_to_fp16, epsilon = var_233_to_fp16, gamma = F0_1_norm2_norm_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("op_264_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_265_cast_fp16 = mul(x = var_261_cast_fp16, y = var_264_cast_fp16)[name = tensor<string, []>("op_265_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_33_cast_fp16 = add(x = var_265_cast_fp16, y = var_258_cast_fp16_1)[name = tensor<string, []>("input_33_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_35_cast_fp16 = leaky_relu(alpha = var_230, x = input_33_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
+            tensor<string, []> input_37_pad_type_0 = const()[name = tensor<string, []>("input_37_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_37_pad_0 = const()[name = tensor<string, []>("input_37_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_37_strides_0 = const()[name = tensor<string, []>("input_37_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_37_dilations_0 = const()[name = tensor<string, []>("input_37_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_37_groups_0 = const()[name = tensor<string, []>("input_37_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 256, 3]> weight_21_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [196608]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4686784))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4883456))), name = tensor<string, []>("weight_21_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 256, 3])];
+            tensor<fp16, [256]> F0_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("F0_2_conv1_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4884032)))];
+            tensor<fp16, [1, 256, ?]> input_37_cast_fp16 = conv(bias = F0_2_conv1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = weight_21_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
+            tensor<fp16, [512, 128]> F0_2_norm2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4884608))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4950208))), name = tensor<string, []>("F0_2_norm2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp16, [512]> F0_2_norm2_fc_bias_to_fp16 = const()[name = tensor<string, []>("F0_2_norm2_fc_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4950784)))];
+            tensor<fp16, [1, 512]> linear_5_cast_fp16 = linear(bias = F0_2_norm2_fc_bias_to_fp16, weight = F0_2_norm2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_5_cast_fp16")];
+            tensor<int32, [3]> var_282 = const()[name = tensor<string, []>("op_282"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp16, [1, 512, 1]> h_23_cast_fp16 = reshape(shape = var_282, x = linear_5_cast_fp16)[name = tensor<string, []>("h_23_cast_fp16")];
+            tensor<int32, [2]> var_284_split_sizes_0 = const()[name = tensor<string, []>("op_284_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_284_axis_0 = const()[name = tensor<string, []>("op_284_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> var_284_cast_fp16_0, tensor<fp16, [1, 256, 1]> var_284_cast_fp16_1 = split(axis = var_284_axis_0, split_sizes = var_284_split_sizes_0, x = h_23_cast_fp16)[name = tensor<string, []>("op_284_cast_fp16")];
+            tensor<fp16, []> var_286_promoted_to_fp16 = const()[name = tensor<string, []>("op_286_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 256, 1]> var_287_cast_fp16 = add(x = var_284_cast_fp16_0, y = var_286_promoted_to_fp16)[name = tensor<string, []>("op_287_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_290_cast_fp16 = instance_norm(beta = F0_1_norm2_norm_bias_to_fp16, epsilon = var_233_to_fp16, gamma = F0_1_norm2_norm_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("op_290_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_291_cast_fp16 = mul(x = var_287_cast_fp16, y = var_290_cast_fp16)[name = tensor<string, []>("op_291_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_39_cast_fp16 = add(x = var_291_cast_fp16, y = var_284_cast_fp16_1)[name = tensor<string, []>("input_39_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_41_cast_fp16 = leaky_relu(alpha = var_230, x = input_39_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
+            tensor<string, []> out_5_pad_type_0 = const()[name = tensor<string, []>("out_5_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> out_5_pad_0 = const()[name = tensor<string, []>("out_5_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> out_5_strides_0 = const()[name = tensor<string, []>("out_5_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> out_5_dilations_0 = const()[name = tensor<string, []>("out_5_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> out_5_groups_0 = const()[name = tensor<string, []>("out_5_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 256, 3]> weight_25_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [196608]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4951872))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5148544))), name = tensor<string, []>("weight_25_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 256, 3])];
+            tensor<fp16, [256]> F0_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("F0_2_conv2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5149120)))];
+            tensor<fp16, [1, 256, ?]> out_5_cast_fp16 = conv(bias = F0_2_conv2_bias_to_fp16, dilations = out_5_dilations_0, groups = out_5_groups_0, pad = out_5_pad_0, pad_type = out_5_pad_type_0, strides = out_5_strides_0, weight = weight_25_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_301_cast_fp16 = add(x = out_5_cast_fp16, y = input_31_cast_fp16)[name = tensor<string, []>("op_301_cast_fp16")];
+            tensor<fp16, []> var_302_to_fp16 = const()[name = tensor<string, []>("op_302_to_fp16"), val = tensor<fp16, []>(0x1.6ap-1)];
+            tensor<fp16, [1, 256, ?]> input_43_cast_fp16 = mul(x = var_301_cast_fp16, y = var_302_to_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
+            tensor<string, []> var_314_pad_type_0 = const()[name = tensor<string, []>("op_314_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_314_strides_0 = const()[name = tensor<string, []>("op_314_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_314_pad_0 = const()[name = tensor<string, []>("op_314_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_314_dilations_0 = const()[name = tensor<string, []>("op_314_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_314_groups_0 = const()[name = tensor<string, []>("op_314_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> F0_proj_weight_to_fp16 = const()[name = tensor<string, []>("F0_proj_weight_to_fp16"), val = tensor<fp16, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5149696)))];
+            tensor<fp16, [1]> F0_proj_bias_to_fp16 = const()[name = tensor<string, []>("F0_proj_bias_to_fp16"), val = tensor<fp16, [1]>([0x1.f08p-3])];
+            tensor<fp16, [1, 1, ?]> var_314_cast_fp16 = conv(bias = F0_proj_bias_to_fp16, dilations = var_314_dilations_0, groups = var_314_groups_0, pad = var_314_pad_0, pad_type = var_314_pad_type_0, strides = var_314_strides_0, weight = F0_proj_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("op_314_cast_fp16")];
+            tensor<int32, [1]> var_316_axes_0 = const()[name = tensor<string, []>("op_316_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, ?]> F0 = squeeze(axes = var_316_axes_0, x = var_314_cast_fp16)[name = tensor<string, []>("op_316_cast_fp16")];
+            tensor<fp32, []> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp16, [1024, 128]> N_0_norm1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5150272))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281408))), name = tensor<string, []>("N_0_norm1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> N_0_norm1_fc_bias_to_fp16 = const()[name = tensor<string, []>("N_0_norm1_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281984)))];
+            tensor<fp16, [1, 1024]> linear_6_cast_fp16 = linear(bias = N_0_norm1_fc_bias_to_fp16, weight = N_0_norm1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_6_cast_fp16")];
+            tensor<int32, [3]> var_347 = const()[name = tensor<string, []>("op_347"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_27_cast_fp16 = reshape(shape = var_347, x = linear_6_cast_fp16)[name = tensor<string, []>("h_27_cast_fp16")];
+            tensor<int32, [2]> var_349_split_sizes_0 = const()[name = tensor<string, []>("op_349_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_349_axis_0 = const()[name = tensor<string, []>("op_349_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_349_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_349_cast_fp16_1 = split(axis = var_349_axis_0, split_sizes = var_349_split_sizes_0, x = h_27_cast_fp16)[name = tensor<string, []>("op_349_cast_fp16")];
+            tensor<fp16, []> var_351_promoted_to_fp16 = const()[name = tensor<string, []>("op_351_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 512, 1]> var_352_cast_fp16 = add(x = var_349_cast_fp16_0, y = var_351_promoted_to_fp16)[name = tensor<string, []>("op_352_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_356_cast_fp16 = mul(x = var_352_cast_fp16, y = var_87_cast_fp16)[name = tensor<string, []>("op_356_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_47_cast_fp16 = add(x = var_356_cast_fp16, y = var_349_cast_fp16_1)[name = tensor<string, []>("input_47_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_49_cast_fp16 = leaky_relu(alpha = var_321, x = input_47_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
+            tensor<string, []> input_51_pad_type_0 = const()[name = tensor<string, []>("input_51_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_51_pad_0 = const()[name = tensor<string, []>("input_51_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_51_strides_0 = const()[name = tensor<string, []>("input_51_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_51_dilations_0 = const()[name = tensor<string, []>("input_51_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_51_groups_0 = const()[name = tensor<string, []>("input_51_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 3]> weight_31_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5284096))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6070592))), name = tensor<string, []>("weight_31_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 3])];
+            tensor<fp16, [512]> N_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("N_0_conv1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6071168)))];
+            tensor<fp16, [1, 512, ?]> input_51_cast_fp16 = conv(bias = N_0_conv1_bias_to_fp16, dilations = input_51_dilations_0, groups = input_51_groups_0, pad = input_51_pad_0, pad_type = input_51_pad_type_0, strides = input_51_strides_0, weight = weight_31_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
+            tensor<fp16, [1024, 128]> N_0_norm2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6072256))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6203392))), name = tensor<string, []>("N_0_norm2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> N_0_norm2_fc_bias_to_fp16 = const()[name = tensor<string, []>("N_0_norm2_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6203968)))];
+            tensor<fp16, [1, 1024]> linear_7_cast_fp16 = linear(bias = N_0_norm2_fc_bias_to_fp16, weight = N_0_norm2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_7_cast_fp16")];
+            tensor<int32, [3]> var_373 = const()[name = tensor<string, []>("op_373"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_31_cast_fp16 = reshape(shape = var_373, x = linear_7_cast_fp16)[name = tensor<string, []>("h_31_cast_fp16")];
+            tensor<int32, [2]> var_375_split_sizes_0 = const()[name = tensor<string, []>("op_375_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_375_axis_0 = const()[name = tensor<string, []>("op_375_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_375_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_375_cast_fp16_1 = split(axis = var_375_axis_0, split_sizes = var_375_split_sizes_0, x = h_31_cast_fp16)[name = tensor<string, []>("op_375_cast_fp16")];
+            tensor<fp16, []> var_377_promoted_to_fp16 = const()[name = tensor<string, []>("op_377_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 512, 1]> var_378_cast_fp16 = add(x = var_375_cast_fp16_0, y = var_377_promoted_to_fp16)[name = tensor<string, []>("op_378_cast_fp16")];
+            tensor<fp16, []> var_324_to_fp16 = const()[name = tensor<string, []>("op_324_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 512, ?]> var_381_cast_fp16 = instance_norm(beta = F0_0_norm1_norm_bias_to_fp16, epsilon = var_324_to_fp16, gamma = F0_0_norm1_norm_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("op_381_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_382_cast_fp16 = mul(x = var_378_cast_fp16, y = var_381_cast_fp16)[name = tensor<string, []>("op_382_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_53_cast_fp16 = add(x = var_382_cast_fp16, y = var_375_cast_fp16_1)[name = tensor<string, []>("input_53_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_55_cast_fp16 = leaky_relu(alpha = var_321, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
+            tensor<string, []> out_7_pad_type_0 = const()[name = tensor<string, []>("out_7_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> out_7_pad_0 = const()[name = tensor<string, []>("out_7_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> out_7_strides_0 = const()[name = tensor<string, []>("out_7_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> out_7_dilations_0 = const()[name = tensor<string, []>("out_7_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> out_7_groups_0 = const()[name = tensor<string, []>("out_7_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 3]> weight_35_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6206080))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6992576))), name = tensor<string, []>("weight_35_to_fp16_palettized"), shape = tensor<uint32, [3]>([512, 512, 3])];
+            tensor<fp16, [512]> N_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("N_0_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6993152)))];
+            tensor<fp16, [1, 512, ?]> out_7_cast_fp16 = conv(bias = N_0_conv2_bias_to_fp16, dilations = out_7_dilations_0, groups = out_7_groups_0, pad = out_7_pad_0, pad_type = out_7_pad_type_0, strides = out_7_strides_0, weight = weight_35_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_392_cast_fp16 = add(x = out_7_cast_fp16, y = transpose_0)[name = tensor<string, []>("op_392_cast_fp16")];
+            tensor<fp16, []> var_393_to_fp16 = const()[name = tensor<string, []>("op_393_to_fp16"), val = tensor<fp16, []>(0x1.6ap-1)];
+            tensor<fp16, [1, 512, ?]> input_57_cast_fp16 = mul(x = var_392_cast_fp16, y = var_393_to_fp16)[name = tensor<string, []>("input_57_cast_fp16")];
+            tensor<fp32, []> var_398 = const()[name = tensor<string, []>("op_398"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp16, [1024, 128]> N_1_norm1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6994240))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7125376))), name = tensor<string, []>("N_1_norm1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 128])];
+            tensor<fp16, [1024]> N_1_norm1_fc_bias_to_fp16 = const()[name = tensor<string, []>("N_1_norm1_fc_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7125952)))];
+            tensor<fp16, [1, 1024]> linear_8_cast_fp16 = linear(bias = N_1_norm1_fc_bias_to_fp16, weight = N_1_norm1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_8_cast_fp16")];
+            tensor<int32, [3]> var_434 = const()[name = tensor<string, []>("op_434"), val = tensor<int32, [3]>([1, 1024, 1])];
+            tensor<fp16, [1, 1024, 1]> h_35_cast_fp16 = reshape(shape = var_434, x = linear_8_cast_fp16)[name = tensor<string, []>("h_35_cast_fp16")];
+            tensor<int32, [2]> var_436_split_sizes_0 = const()[name = tensor<string, []>("op_436_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
+            tensor<int32, []> var_436_axis_0 = const()[name = tensor<string, []>("op_436_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, 1]> var_436_cast_fp16_0, tensor<fp16, [1, 512, 1]> var_436_cast_fp16_1 = split(axis = var_436_axis_0, split_sizes = var_436_split_sizes_0, x = h_35_cast_fp16)[name = tensor<string, []>("op_436_cast_fp16")];
+            tensor<fp16, []> var_438_promoted_to_fp16 = const()[name = tensor<string, []>("op_438_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 512, 1]> var_439_cast_fp16 = add(x = var_436_cast_fp16_0, y = var_438_promoted_to_fp16)[name = tensor<string, []>("op_439_cast_fp16")];
+            tensor<fp16, []> var_402_to_fp16 = const()[name = tensor<string, []>("op_402_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 512, ?]> var_442_cast_fp16 = instance_norm(beta = F0_0_norm1_norm_bias_to_fp16, epsilon = var_402_to_fp16, gamma = F0_0_norm1_norm_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("op_442_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_443_cast_fp16 = mul(x = var_439_cast_fp16, y = var_442_cast_fp16)[name = tensor<string, []>("op_443_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_59_cast_fp16 = add(x = var_443_cast_fp16, y = var_436_cast_fp16_1)[name = tensor<string, []>("input_59_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> input_61_cast_fp16 = leaky_relu(alpha = var_398, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
+            tensor<string, []> conv_transpose_1_pad_type_0 = const()[name = tensor<string, []>("conv_transpose_1_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> conv_transpose_1_pad_0 = const()[name = tensor<string, []>("conv_transpose_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_transpose_1_strides_0 = const()[name = tensor<string, []>("conv_transpose_1_strides_0"), val = tensor<int32, [1]>([2])];
+            tensor<int32, []> conv_transpose_1_groups_0 = const()[name = tensor<string, []>("conv_transpose_1_groups_0"), val = tensor<int32, []>(512)];
+            tensor<int32, [1]> conv_transpose_1_dilations_0 = const()[name = tensor<string, []>("conv_transpose_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [512, 1, 3]> var_446_to_fp16 = const()[name = tensor<string, []>("op_446_to_fp16"), val = tensor<fp16, [512, 1, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7128064)))];
+            tensor<fp16, [512]> N_1_pool_bias_to_fp16 = const()[name = tensor<string, []>("N_1_pool_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7131200)))];
+            tensor<fp16, [1, 512, ?]> conv_transpose_1_cast_fp16 = conv_transpose(bias = N_1_pool_bias_to_fp16, dilations = conv_transpose_1_dilations_0, groups = conv_transpose_1_groups_0, pad = conv_transpose_1_pad_0, pad_type = conv_transpose_1_pad_type_0, strides = conv_transpose_1_strides_0, weight = var_446_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("conv_transpose_1_cast_fp16")];
+            tensor<int32, [3]> input_63_begin_0 = const()[name = tensor<string, []>("input_63_begin_0"), val = tensor<int32, [3]>([0, 0, 1])];
+            tensor<int32, [3]> input_63_end_0 = const()[name = tensor<string, []>("input_63_end_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<bool, [3]> input_63_begin_mask_0 = const()[name = tensor<string, []>("input_63_begin_mask_0"), val = tensor<bool, [3]>([true, true, false])];
+            tensor<bool, [3]> input_63_end_mask_0 = const()[name = tensor<string, []>("input_63_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
+            tensor<fp16, [1, 512, ?]> input_63_cast_fp16 = slice_by_index(begin = input_63_begin_0, begin_mask = input_63_begin_mask_0, end = input_63_end_0, end_mask = input_63_end_mask_0, x = conv_transpose_1_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
+            tensor<string, []> input_65_pad_type_0 = const()[name = tensor<string, []>("input_65_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_65_pad_0 = const()[name = tensor<string, []>("input_65_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_65_strides_0 = const()[name = tensor<string, []>("input_65_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_65_dilations_0 = const()[name = tensor<string, []>("input_65_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_65_groups_0 = const()[name = tensor<string, []>("input_65_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 512, 3]> weight_39_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [393216]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7132288))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7525568))), name = tensor<string, []>("weight_39_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 512, 3])];
+            tensor<fp16, [256]> N_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("N_1_conv1_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7526144)))];
+            tensor<fp16, [1, 256, ?]> input_65_cast_fp16 = conv(bias = N_1_conv1_bias_to_fp16, dilations = input_65_dilations_0, groups = input_65_groups_0, pad = input_65_pad_0, pad_type = input_65_pad_type_0, strides = input_65_strides_0, weight = weight_39_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
+            tensor<fp16, [512, 128]> N_1_norm2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7526720))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7592320))), name = tensor<string, []>("N_1_norm2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp16, [512]> N_1_norm2_fc_bias_to_fp16 = const()[name = tensor<string, []>("N_1_norm2_fc_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7592896)))];
+            tensor<fp16, [1, 512]> linear_9_cast_fp16 = linear(bias = N_1_norm2_fc_bias_to_fp16, weight = N_1_norm2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_9_cast_fp16")];
+            tensor<int32, [3]> var_467 = const()[name = tensor<string, []>("op_467"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp16, [1, 512, 1]> h_39_cast_fp16 = reshape(shape = var_467, x = linear_9_cast_fp16)[name = tensor<string, []>("h_39_cast_fp16")];
+            tensor<int32, [2]> var_469_split_sizes_0 = const()[name = tensor<string, []>("op_469_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_469_axis_0 = const()[name = tensor<string, []>("op_469_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> var_469_cast_fp16_0, tensor<fp16, [1, 256, 1]> var_469_cast_fp16_1 = split(axis = var_469_axis_0, split_sizes = var_469_split_sizes_0, x = h_39_cast_fp16)[name = tensor<string, []>("op_469_cast_fp16")];
+            tensor<fp16, []> var_471_promoted_to_fp16 = const()[name = tensor<string, []>("op_471_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 256, 1]> var_472_cast_fp16 = add(x = var_469_cast_fp16_0, y = var_471_promoted_to_fp16)[name = tensor<string, []>("op_472_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_475_cast_fp16 = instance_norm(beta = F0_1_norm2_norm_bias_to_fp16, epsilon = var_402_to_fp16, gamma = F0_1_norm2_norm_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("op_475_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_476_cast_fp16 = mul(x = var_472_cast_fp16, y = var_475_cast_fp16)[name = tensor<string, []>("op_476_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_67_cast_fp16 = add(x = var_476_cast_fp16, y = var_469_cast_fp16_1)[name = tensor<string, []>("input_67_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_69_cast_fp16 = leaky_relu(alpha = var_398, x = input_67_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
+            tensor<string, []> out_9_pad_type_0 = const()[name = tensor<string, []>("out_9_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> out_9_pad_0 = const()[name = tensor<string, []>("out_9_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> out_9_strides_0 = const()[name = tensor<string, []>("out_9_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> out_9_dilations_0 = const()[name = tensor<string, []>("out_9_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> out_9_groups_0 = const()[name = tensor<string, []>("out_9_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 256, 3]> weight_43_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [196608]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7593984))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7790656))), name = tensor<string, []>("weight_43_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 256, 3])];
+            tensor<fp16, [256]> N_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("N_1_conv2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7791232)))];
+            tensor<fp16, [1, 256, ?]> out_9_cast_fp16 = conv(bias = N_1_conv2_bias_to_fp16, dilations = out_9_dilations_0, groups = out_9_groups_0, pad = out_9_pad_0, pad_type = out_9_pad_type_0, strides = out_9_strides_0, weight = weight_43_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")];
+            tensor<int32, [1]> expand_dims_1_axes_0 = const()[name = tensor<string, []>("expand_dims_1_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp16, [1, 512, ?, 1]> expand_dims_1_cast_fp16 = expand_dims(axes = expand_dims_1_axes_0, x = input_57_cast_fp16)[name = tensor<string, []>("expand_dims_1_cast_fp16")];
+            tensor<int32, []> upsample_nearest_neighbor_1_scale_factor_height_0 = const()[name = tensor<string, []>("upsample_nearest_neighbor_1_scale_factor_height_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> upsample_nearest_neighbor_1_scale_factor_width_0 = const()[name = tensor<string, []>("upsample_nearest_neighbor_1_scale_factor_width_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 512, ?, 1]> upsample_nearest_neighbor_1_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = upsample_nearest_neighbor_1_scale_factor_height_0, scale_factor_width = upsample_nearest_neighbor_1_scale_factor_width_0, x = expand_dims_1_cast_fp16)[name = tensor<string, []>("upsample_nearest_neighbor_1_cast_fp16")];
+            tensor<int32, [1]> input_71_axes_0 = const()[name = tensor<string, []>("input_71_axes_0"), val = tensor<int32, [1]>([3])];
+            tensor<fp16, [1, 512, ?]> input_71_cast_fp16 = squeeze(axes = input_71_axes_0, x = upsample_nearest_neighbor_1_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
+            tensor<string, []> var_493_pad_type_0 = const()[name = tensor<string, []>("op_493_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_493_strides_0 = const()[name = tensor<string, []>("op_493_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_493_pad_0 = const()[name = tensor<string, []>("op_493_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_493_dilations_0 = const()[name = tensor<string, []>("op_493_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_493_groups_0 = const()[name = tensor<string, []>("op_493_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 512, 1]> weight_45_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [131072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7791808))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7922944))), name = tensor<string, []>("weight_45_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 512, 1])];
+            tensor<fp16, [1, 256, ?]> var_493_cast_fp16 = conv(dilations = var_493_dilations_0, groups = var_493_groups_0, pad = var_493_pad_0, pad_type = var_493_pad_type_0, strides = var_493_strides_0, weight = weight_45_to_fp16_palettized, x = input_71_cast_fp16)[name = tensor<string, []>("op_493_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_494_cast_fp16 = add(x = out_9_cast_fp16, y = var_493_cast_fp16)[name = tensor<string, []>("op_494_cast_fp16")];
+            tensor<fp16, []> var_495_to_fp16 = const()[name = tensor<string, []>("op_495_to_fp16"), val = tensor<fp16, []>(0x1.6ap-1)];
+            tensor<fp16, [1, 256, ?]> input_73_cast_fp16 = mul(x = var_494_cast_fp16, y = var_495_to_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
+            tensor<fp32, []> var_498 = const()[name = tensor<string, []>("op_498"), val = tensor<fp32, []>(0x1.99999ap-3)];
+            tensor<fp16, [512, 128]> N_2_norm1_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7923520))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7989120))), name = tensor<string, []>("N_2_norm1_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp16, [512]> N_2_norm1_fc_bias_to_fp16 = const()[name = tensor<string, []>("N_2_norm1_fc_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7989696)))];
+            tensor<fp16, [1, 512]> linear_10_cast_fp16 = linear(bias = N_2_norm1_fc_bias_to_fp16, weight = N_2_norm1_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_10_cast_fp16")];
+            tensor<int32, [3]> var_524 = const()[name = tensor<string, []>("op_524"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp16, [1, 512, 1]> h_43_cast_fp16 = reshape(shape = var_524, x = linear_10_cast_fp16)[name = tensor<string, []>("h_43_cast_fp16")];
+            tensor<int32, [2]> var_526_split_sizes_0 = const()[name = tensor<string, []>("op_526_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_526_axis_0 = const()[name = tensor<string, []>("op_526_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> var_526_cast_fp16_0, tensor<fp16, [1, 256, 1]> var_526_cast_fp16_1 = split(axis = var_526_axis_0, split_sizes = var_526_split_sizes_0, x = h_43_cast_fp16)[name = tensor<string, []>("op_526_cast_fp16")];
+            tensor<fp16, []> var_528_promoted_to_fp16 = const()[name = tensor<string, []>("op_528_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 256, 1]> var_529_cast_fp16 = add(x = var_526_cast_fp16_0, y = var_528_promoted_to_fp16)[name = tensor<string, []>("op_529_cast_fp16")];
+            tensor<fp16, []> var_501_to_fp16 = const()[name = tensor<string, []>("op_501_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 256, ?]> var_532_cast_fp16 = instance_norm(beta = F0_1_norm2_norm_bias_to_fp16, epsilon = var_501_to_fp16, gamma = F0_1_norm2_norm_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("op_532_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_533_cast_fp16 = mul(x = var_529_cast_fp16, y = var_532_cast_fp16)[name = tensor<string, []>("op_533_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_75_cast_fp16 = add(x = var_533_cast_fp16, y = var_526_cast_fp16_1)[name = tensor<string, []>("input_75_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_77_cast_fp16 = leaky_relu(alpha = var_498, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
+            tensor<string, []> input_79_pad_type_0 = const()[name = tensor<string, []>("input_79_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_79_pad_0 = const()[name = tensor<string, []>("input_79_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_79_strides_0 = const()[name = tensor<string, []>("input_79_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_79_dilations_0 = const()[name = tensor<string, []>("input_79_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_79_groups_0 = const()[name = tensor<string, []>("input_79_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 256, 3]> weight_49_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [196608]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7990784))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8187456))), name = tensor<string, []>("weight_49_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 256, 3])];
+            tensor<fp16, [256]> N_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("N_2_conv1_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8188032)))];
+            tensor<fp16, [1, 256, ?]> input_79_cast_fp16 = conv(bias = N_2_conv1_bias_to_fp16, dilations = input_79_dilations_0, groups = input_79_groups_0, pad = input_79_pad_0, pad_type = input_79_pad_type_0, strides = input_79_strides_0, weight = weight_49_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")];
+            tensor<fp16, [512, 128]> N_2_norm2_fc_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [65536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8188608))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8254208))), name = tensor<string, []>("N_2_norm2_fc_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([512, 128])];
+            tensor<fp16, [512]> N_2_norm2_fc_bias_to_fp16 = const()[name = tensor<string, []>("N_2_norm2_fc_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8254784)))];
+            tensor<fp16, [1, 512]> linear_11_cast_fp16 = linear(bias = N_2_norm2_fc_bias_to_fp16, weight = N_2_norm2_fc_weight_to_fp16_palettized, x = style_s)[name = tensor<string, []>("linear_11_cast_fp16")];
+            tensor<int32, [3]> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<int32, [3]>([1, 512, 1])];
+            tensor<fp16, [1, 512, 1]> h_cast_fp16 = reshape(shape = var_550, x = linear_11_cast_fp16)[name = tensor<string, []>("h_cast_fp16")];
+            tensor<int32, [2]> var_552_split_sizes_0 = const()[name = tensor<string, []>("op_552_split_sizes_0"), val = tensor<int32, [2]>([256, 256])];
+            tensor<int32, []> var_552_axis_0 = const()[name = tensor<string, []>("op_552_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> var_552_cast_fp16_0, tensor<fp16, [1, 256, 1]> var_552_cast_fp16_1 = split(axis = var_552_axis_0, split_sizes = var_552_split_sizes_0, x = h_cast_fp16)[name = tensor<string, []>("op_552_cast_fp16")];
+            tensor<fp16, []> var_554_promoted_to_fp16 = const()[name = tensor<string, []>("op_554_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1, 256, 1]> var_555_cast_fp16 = add(x = var_552_cast_fp16_0, y = var_554_promoted_to_fp16)[name = tensor<string, []>("op_555_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_558_cast_fp16 = instance_norm(beta = F0_1_norm2_norm_bias_to_fp16, epsilon = var_501_to_fp16, gamma = F0_1_norm2_norm_weight_to_fp16, x = input_79_cast_fp16)[name = tensor<string, []>("op_558_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_559_cast_fp16 = mul(x = var_555_cast_fp16, y = var_558_cast_fp16)[name = tensor<string, []>("op_559_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_81_cast_fp16 = add(x = var_559_cast_fp16, y = var_552_cast_fp16_1)[name = tensor<string, []>("input_81_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> input_83_cast_fp16 = leaky_relu(alpha = var_498, x = input_81_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
+            tensor<string, []> out_pad_type_0 = const()[name = tensor<string, []>("out_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> out_pad_0 = const()[name = tensor<string, []>("out_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> out_strides_0 = const()[name = tensor<string, []>("out_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> out_dilations_0 = const()[name = tensor<string, []>("out_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> out_groups_0 = const()[name = tensor<string, []>("out_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [256, 256, 3]> weight_53_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [196608]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8255872))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8452544))), name = tensor<string, []>("weight_53_to_fp16_palettized"), shape = tensor<uint32, [3]>([256, 256, 3])];
+            tensor<fp16, [256]> N_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("N_2_conv2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8453120)))];
+            tensor<fp16, [1, 256, ?]> out_cast_fp16 = conv(bias = N_2_conv2_bias_to_fp16, dilations = out_dilations_0, groups = out_groups_0, pad = out_pad_0, pad_type = out_pad_type_0, strides = out_strides_0, weight = weight_53_to_fp16_palettized, x = input_83_cast_fp16)[name = tensor<string, []>("out_cast_fp16")];
+            tensor<fp16, [1, 256, ?]> var_569_cast_fp16 = add(x = out_cast_fp16, y = input_73_cast_fp16)[name = tensor<string, []>("op_569_cast_fp16")];
+            tensor<fp16, []> var_570_to_fp16 = const()[name = tensor<string, []>("op_570_to_fp16"), val = tensor<fp16, []>(0x1.6ap-1)];
+            tensor<fp16, [1, 256, ?]> input_cast_fp16 = mul(x = var_569_cast_fp16, y = var_570_to_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<string, []> var_582_pad_type_0 = const()[name = tensor<string, []>("op_582_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_582_strides_0 = const()[name = tensor<string, []>("op_582_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_582_pad_0 = const()[name = tensor<string, []>("op_582_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_582_dilations_0 = const()[name = tensor<string, []>("op_582_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_582_groups_0 = const()[name = tensor<string, []>("op_582_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 256, 1]> N_proj_weight_to_fp16 = const()[name = tensor<string, []>("N_proj_weight_to_fp16"), val = tensor<fp16, [1, 256, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8453696)))];
+            tensor<fp16, [1]> N_proj_bias_to_fp16 = const()[name = tensor<string, []>("N_proj_bias_to_fp16"), val = tensor<fp16, [1]>([0x1.3fp-5])];
+            tensor<fp16, [1, 1, ?]> var_582_cast_fp16 = conv(bias = N_proj_bias_to_fp16, dilations = var_582_dilations_0, groups = var_582_groups_0, pad = var_582_pad_0, pad_type = var_582_pad_type_0, strides = var_582_strides_0, weight = N_proj_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("op_582_cast_fp16")];
+            tensor<int32, [1]> var_584_axes_0 = const()[name = tensor<string, []>("op_584_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, ?]> N = squeeze(axes = var_584_axes_0, x = var_582_cast_fp16)[name = tensor<string, []>("op_584_cast_fp16")];
+        } -> (F0, N);
+}

ANE/ANE-zh/KokoroProsody.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d7229d31a47e1d5c054c24b7aed8ce0df20460523472a55ff998f5939a75cf8
+size 8454272

ANE/ANE-zh/KokoroProsody.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:801253b6ccdd10a0e8c991f66a34c6bbd05f98a1115c11836a5aa9778b7bed93
+size 65723

ANE/ANE-zh/KokoroProsody.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d7229d31a47e1d5c054c24b7aed8ce0df20460523472a55ff998f5939a75cf8
+size 8454272

ANE/ANE-zh/KokoroProsody.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "A584E1D6-B88E-412B-83F8-60F488E36916": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "C7F601FE-E520-4ADA-B688-25A6DC60355C": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "A584E1D6-B88E-412B-83F8-60F488E36916"
+}

ANE/ANE-zh/KokoroTail.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbffea509dfcba72fae7a9dc7ae424e19f8af08eabc70360fe30fd7c1de09151
+size 243

ANE/ANE-zh/KokoroTail.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94e611a84f91c7b135b031a0f978cc47b0edab42912e68a86c3f3e78f9edf6a0
+size 392

ANE/ANE-zh/KokoroTail.mlmodelc/metadata.json

@@ -0,0 +1,70 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float32",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "audio",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.sin" : 2,
+      "Ios17.convTranspose" : 2,
+      "Ios17.conv" : 1,
+      "Ios17.cos" : 1,
+      "Ios17.sliceByIndex" : 3,
+      "Ios17.mul" : 2,
+      "Ios17.sub" : 1,
+      "Ios17.exp" : 1
+    },
+    "computePrecision" : "Mixed (Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-05-03",
+      "com.github.apple.coremltools.source" : "torch==2.11.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 128 × 100...240001",
+        "shapeRange" : "[[1, 1], [128, 128], [100, 240001]]",
+        "formattedType" : "MultiArray (Float32 1 × 128 × 15961)",
+        "type" : "MultiArray",
+        "shape" : "[1, 128, 15961]",
+        "name" : "x_pre",
+        "shortDescription" : ""
+      }
+    ],
+    "generatedClassName" : "KokoroTail",
+    "method" : "predict"
+  }
+]

ANE/ANE-zh/KokoroTail.mlmodelc/model.mil

@@ -0,0 +1,47 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.11.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios17>(tensor<fp32, [1, 128, ?]> x_pre) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"x_pre", [1, 128, 15961]}}), ("RangeDims", {{"x_pre", [[1, 1], [128, 128], [100, 240001]]}})))] {
+            tensor<fp32, [22]> conv_post_bias = const()[name = tensor<string, []>("conv_post_bias"), val = tensor<fp32, [22]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp32, [11, 1, 20]> stft_deconv_real_weight = const()[name = tensor<string, []>("stft_deconv_real_weight"), val = tensor<fp32, [11, 1, 20]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256)))];
+            tensor<fp32, [11, 1, 20]> stft_deconv_imag_weight = const()[name = tensor<string, []>("stft_deconv_imag_weight"), val = tensor<fp32, [11, 1, 20]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1216)))];
+            tensor<fp32, [22, 128, 7]> weight_1 = const()[name = tensor<string, []>("weight_1"), val = tensor<fp32, [22, 128, 7]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2176)))];
+            tensor<string, []> x_pad_type_0 = const()[name = tensor<string, []>("x_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_pad_0 = const()[name = tensor<string, []>("x_pad_0"), val = tensor<int32, [2]>([3, 3])];
+            tensor<int32, [1]> x_strides_0 = const()[name = tensor<string, []>("x_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> x_dilations_0 = const()[name = tensor<string, []>("x_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_groups_0 = const()[name = tensor<string, []>("x_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 22, ?]> x = conv(bias = conv_post_bias, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = weight_1, x = x_pre)[name = tensor<string, []>("x")];
+            tensor<int32, [3]> var_33_begin_0 = const()[name = tensor<string, []>("op_33_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<int32, [3]> var_33_end_0 = const()[name = tensor<string, []>("op_33_end_0"), val = tensor<int32, [3]>([1, 11, 0])];
+            tensor<bool, [3]> var_33_end_mask_0 = const()[name = tensor<string, []>("op_33_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
+            tensor<fp32, [1, 11, ?]> var_33 = slice_by_index(begin = var_33_begin_0, end = var_33_end_0, end_mask = var_33_end_mask_0, x = x)[name = tensor<string, []>("op_33")];
+            tensor<fp32, [1, 11, ?]> magnitude = exp(x = var_33)[name = tensor<string, []>("magnitude")];
+            tensor<int32, [3]> var_49_begin_0 = const()[name = tensor<string, []>("op_49_begin_0"), val = tensor<int32, [3]>([0, 11, 0])];
+            tensor<int32, [3]> var_49_end_0 = const()[name = tensor<string, []>("op_49_end_0"), val = tensor<int32, [3]>([1, 22, 0])];
+            tensor<bool, [3]> var_49_end_mask_0 = const()[name = tensor<string, []>("op_49_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
+            tensor<fp32, [1, 11, ?]> var_49 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = x)[name = tensor<string, []>("op_49")];
+            tensor<fp32, [1, 11, ?]> phase = sin(x = var_49)[name = tensor<string, []>("phase")];
+            tensor<fp32, [1, 11, ?]> var_56 = cos(x = phase)[name = tensor<string, []>("op_56")];
+            tensor<fp32, [1, 11, ?]> input_1 = mul(x = magnitude, y = var_56)[name = tensor<string, []>("input_1")];
+            tensor<fp32, [1, 11, ?]> var_58 = sin(x = phase)[name = tensor<string, []>("op_58")];
+            tensor<fp32, [1, 11, ?]> input = mul(x = magnitude, y = var_58)[name = tensor<string, []>("input")];
+            tensor<string, []> var_71_pad_type_0 = const()[name = tensor<string, []>("op_71_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_71_strides_0 = const()[name = tensor<string, []>("op_71_strides_0"), val = tensor<int32, [1]>([5])];
+            tensor<int32, [2]> var_71_pad_0 = const()[name = tensor<string, []>("op_71_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_71_dilations_0 = const()[name = tensor<string, []>("op_71_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_71_groups_0 = const()[name = tensor<string, []>("op_71_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 1, ?]> var_71 = conv_transpose(dilations = var_71_dilations_0, groups = var_71_groups_0, pad = var_71_pad_0, pad_type = var_71_pad_type_0, strides = var_71_strides_0, weight = stft_deconv_real_weight, x = input_1)[name = tensor<string, []>("op_71")];
+            tensor<string, []> var_83_pad_type_0 = const()[name = tensor<string, []>("op_83_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_83_strides_0 = const()[name = tensor<string, []>("op_83_strides_0"), val = tensor<int32, [1]>([5])];
+            tensor<int32, [2]> var_83_pad_0 = const()[name = tensor<string, []>("op_83_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_83_dilations_0 = const()[name = tensor<string, []>("op_83_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_83_groups_0 = const()[name = tensor<string, []>("op_83_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 1, ?]> var_83 = conv_transpose(dilations = var_83_dilations_0, groups = var_83_groups_0, pad = var_83_pad_0, pad_type = var_83_pad_type_0, strides = var_83_strides_0, weight = stft_deconv_imag_weight, x = input)[name = tensor<string, []>("op_83")];
+            tensor<fp32, [1, 1, ?]> waveform = sub(x = var_71, y = var_83)[name = tensor<string, []>("waveform")];
+            tensor<int32, [3]> var_90_begin_0 = const()[name = tensor<string, []>("op_90_begin_0"), val = tensor<int32, [3]>([0, 0, 10])];
+            tensor<int32, [3]> var_90_end_0 = const()[name = tensor<string, []>("op_90_end_0"), val = tensor<int32, [3]>([1, 1, -10])];
+            tensor<bool, [3]> var_90_end_mask_0 = const()[name = tensor<string, []>("op_90_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
+            tensor<fp32, [1, 1, ?]> audio = slice_by_index(begin = var_90_begin_0, end = var_90_end_0, end_mask = var_90_end_mask_0, x = waveform)[name = tensor<string, []>("op_90")];
+        } -> (audio);
+}

ANE/ANE-zh/KokoroTail.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1865207df8b7608f3fd443b5a3c744634a8942ccc917b5c8734818d569c0f4eb
+size 81088

ANE/ANE-zh/KokoroTail.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c95acf7775be2c4832b10a81517b4b133dd01b4ed0b54a41f0d7689ffca8526
+size 5964

ANE/ANE-zh/KokoroTail.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1865207df8b7608f3fd443b5a3c744634a8942ccc917b5c8734818d569c0f4eb
+size 81088

ANE/ANE-zh/KokoroTail.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "18BB1DC6-314D-4189-A295-F891E896DE7B": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "E288AD71-C613-4173-B1DC-4CEC042D94B8": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "E288AD71-C613-4173-B1DC-4CEC042D94B8"
+}

ANE/ANE-zh/KokoroVocoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12bf39f5117a2fe2645b0158d75d71a99b328b237f1add55a1511d0e2ee3b456
+size 243

ANE/ANE-zh/KokoroVocoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24ad83fed32d94eff2b3d9d70159140057ebad500b2a7f8fb54b0be0204b1cd2
+size 627