Upload 88 files

compiled/parakeet_ctc_coreml/parakeet_ctc_decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:257f0247edf788b1ee34cc766f13441dc37bbb76e263207d622eceebfde004c1
+size 243

compiled/parakeet_ctc_coreml/parakeet_ctc_decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b42fcc1728ae57657aef0d04be604304285d11b3bbf56c38c105fc498af05b2
+size 463

compiled/parakeet_ctc_coreml/parakeet_ctc_decoder.mlmodelc/metadata.json

@@ -0,0 +1,69 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC decoder head (encoder -> log_probs)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "log_probs",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.transpose" : 1,
+      "Ios18.cast" : 2,
+      "Ios18.conv" : 1,
+      "Ios18.softmax" : 1,
+      "Ios18.log" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 1024 × 1...188",
+        "shapeRange" : "[[1, 1], [1024, 1024], [1, 188]]",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 1)",
+        "type" : "MultiArray",
+        "shape" : "[1, 1024, 1]",
+        "name" : "encoder",
+        "shortDescription" : ""
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_decoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml/parakeet_ctc_decoder.mlmodelc/model.mil

@@ -0,0 +1,24 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.10.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios18>(tensor<fp32, [1, 1024, ?]> encoder) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"encoder", [1, 1024, 1]}}), ("RangeDims", {{"encoder", [[1, 1], [1024, 1024], [1, 188]]}})))] {
+            int32 var_4 = const()[name = string("op_4"), val = int32(-1)];
+            string var_18_pad_type_0 = const()[name = string("op_18_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> var_18_strides_0 = const()[name = string("op_18_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_18_pad_0 = const()[name = string("op_18_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_18_dilations_0 = const()[name = string("op_18_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 var_18_groups_0 = const()[name = string("op_18_groups_0"), val = int32(1)];
+            string encoder_to_fp16_dtype_0 = const()[name = string("encoder_to_fp16_dtype_0"), val = string("fp16")];
+            tensor<fp16, [1025, 1024, 1]> module_decoder_layers_0_weight_to_fp16 = const()[name = string("module_decoder_layers_0_weight_to_fp16"), val = tensor<fp16, [1025, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
+            tensor<fp16, [1025]> module_decoder_layers_0_bias_to_fp16 = const()[name = string("module_decoder_layers_0_bias_to_fp16"), val = tensor<fp16, [1025]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2099328)))];
+            tensor<fp16, [1, 1024, ?]> encoder_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = string("cast_1")];
+            tensor<fp16, [1, 1025, ?]> var_18_cast_fp16 = conv(bias = module_decoder_layers_0_bias_to_fp16, dilations = var_18_dilations_0, groups = var_18_groups_0, pad = var_18_pad_0, pad_type = var_18_pad_type_0, strides = var_18_strides_0, weight = module_decoder_layers_0_weight_to_fp16, x = encoder_to_fp16)[name = string("op_18_cast_fp16")];
+            tensor<int32, [3]> input_perm_0 = const()[name = string("input_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, ?, 1025]> input_cast_fp16 = transpose(perm = input_perm_0, x = var_18_cast_fp16)[name = string("transpose_0")];
+            tensor<fp16, [1, ?, 1025]> out_objects_softmax_cast_fp16 = softmax(axis = var_4, x = input_cast_fp16)[name = string("out_objects_softmax_cast_fp16")];
+            fp32 out_objects_epsilon_0 = const()[name = string("out_objects_epsilon_0"), val = fp32(0x1p-149)];
+            tensor<fp16, [1, ?, 1025]> out_objects_cast_fp16 = log(epsilon = out_objects_epsilon_0, x = out_objects_softmax_cast_fp16)[name = string("out_objects_cast_fp16")];
+            string out_objects_cast_fp16_to_fp32_dtype_0 = const()[name = string("out_objects_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp32, [1, ?, 1025]> log_probs = cast(dtype = out_objects_cast_fp16_to_fp32_dtype_0, x = out_objects_cast_fp16)[name = string("cast_0")];
+        } -> (log_probs);
+}

compiled/parakeet_ctc_coreml/parakeet_ctc_decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4676cb50ed9593fadebd8e5917d728dc437e2858a92695f8225237461dc4e3ea
+size 2101442

compiled/parakeet_ctc_coreml/parakeet_ctc_mel_encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b19f1f034a958dc31d6c6e2049ede86e5d345958dc2a4de9c094f4a592279f9d
+size 243

compiled/parakeet_ctc_coreml/parakeet_ctc_mel_encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fb7247da4f4d4a8168ea123505c35c2098dd2e3830e4a618c42260b6bcbece4
+size 496

compiled/parakeet_ctc_coreml/parakeet_ctc_mel_encoder.mlmodelc/metadata.json

@@ -0,0 +1,117 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC Mel+Encoder (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 188)",
+        "shortDescription" : "",
+        "shape" : "[1, 1024, 188]",
+        "name" : "encoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "encoder_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int32)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Stack" : 1,
+      "Ios18.conv" : 79,
+      "Ios18.mul" : 107,
+      "Ios18.cast" : 12,
+      "Identity" : 1,
+      "Ios18.layerNorm" : 120,
+      "Ios18.log" : 1,
+      "Ios18.equal" : 1,
+      "Ios18.floorDiv" : 4,
+      "Ios16.reduceSum" : 4,
+      "Ios18.logicalNot" : 3,
+      "Ios18.reshape" : 147,
+      "Pad" : 49,
+      "Ios18.concat" : 1,
+      "Ios18.add" : 180,
+      "Ios18.realDiv" : 3,
+      "Ios18.greaterEqual" : 1,
+      "Ios18.notEqual" : 1,
+      "Ios18.relu" : 3,
+      "Ios18.sub" : 7,
+      "Ios18.matmul" : 73,
+      "Ios18.silu" : 72,
+      "Ios18.expandDims" : 25,
+      "Ios18.linear" : 193,
+      "Ios18.sigmoid" : 24,
+      "Ios18.sliceByIndex" : 51,
+      "Ios18.transpose" : 172,
+      "Split" : 24,
+      "Ios18.sqrt" : 1,
+      "Ios18.softmax" : 24,
+      "Ios18.pow" : 2,
+      "Select" : 78,
+      "Ios18.logicalAnd" : 2,
+      "Ios18.less" : 6,
+      "Tile" : 9
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 240000)",
+        "shortDescription" : "",
+        "shape" : "[1, 240000]",
+        "name" : "audio_signal",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "audio_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_mel_encoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml/parakeet_ctc_mel_encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94f8809383b04af30f1f248d47f4cb159fcf58dd6dccb93158c8600fad16f158
+size 1185334464

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08c3096db11abbb0e71805257a62863e41310aa2419750cde0435957990af174
+size 243

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bb8115c6d918526b13db13e228e22dd8010269a545d922d83d6a10ab0dedc42
+size 463

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_decoder.mlmodelc/metadata.json

@@ -0,0 +1,70 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC decoder head (encoder -> log_probs)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "log_probs",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Palettized (4 bits), UInt4)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.transpose" : 1,
+      "Ios18.constexprLutToDense" : 1,
+      "Ios18.cast" : 2,
+      "Ios18.conv" : 1,
+      "Ios18.softmax" : 1,
+      "Ios18.log" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 1024 × 1...188",
+        "shapeRange" : "[[1, 1], [1024, 1024], [1, 188]]",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 1)",
+        "type" : "MultiArray",
+        "shape" : "[1, 1024, 1]",
+        "name" : "encoder",
+        "shortDescription" : ""
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_decoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_decoder.mlmodelc/model.mil

@@ -0,0 +1,24 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
+{
+    func main<ios18>(tensor<fp32, [1, 1024, ?]> encoder) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"encoder", [1, 1024, 1]}}), ("RangeDims", {{"encoder", [[1, 1], [1024, 1024], [1, 188]]}})))] {
+            int32 var_4 = const()[name = string("op_4"), val = int32(-1)];
+            string var_18_pad_type_0 = const()[name = string("op_18_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> var_18_strides_0 = const()[name = string("op_18_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_18_pad_0 = const()[name = string("op_18_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_18_dilations_0 = const()[name = string("op_18_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 var_18_groups_0 = const()[name = string("op_18_groups_0"), val = int32(1)];
+            string encoder_to_fp16_dtype_0 = const()[name = string("encoder_to_fp16_dtype_0"), val = string("fp16")];
+            tensor<fp16, [1025, 1024, 1]> module_decoder_layers_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1025, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(524928))))[name = string("module_decoder_layers_0_weight_to_fp16_palettized")];
+            tensor<fp16, [1025]> module_decoder_layers_0_bias_to_fp16 = const()[name = string("module_decoder_layers_0_bias_to_fp16"), val = tensor<fp16, [1025]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(525056)))];
+            tensor<fp16, [1, 1024, ?]> encoder_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = string("cast_1")];
+            tensor<fp16, [1, 1025, ?]> var_18_cast_fp16 = conv(bias = module_decoder_layers_0_bias_to_fp16, dilations = var_18_dilations_0, groups = var_18_groups_0, pad = var_18_pad_0, pad_type = var_18_pad_type_0, strides = var_18_strides_0, weight = module_decoder_layers_0_weight_to_fp16_palettized, x = encoder_to_fp16)[name = string("op_18_cast_fp16")];
+            tensor<int32, [3]> input_perm_0 = const()[name = string("input_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, ?, 1025]> input_cast_fp16 = transpose(perm = input_perm_0, x = var_18_cast_fp16)[name = string("transpose_0")];
+            tensor<fp16, [1, ?, 1025]> out_objects_softmax_cast_fp16 = softmax(axis = var_4, x = input_cast_fp16)[name = string("out_objects_softmax_cast_fp16")];
+            fp32 out_objects_epsilon_0 = const()[name = string("out_objects_epsilon_0"), val = fp32(0x1p-149)];
+            tensor<fp16, [1, ?, 1025]> out_objects_cast_fp16 = log(epsilon = out_objects_epsilon_0, x = out_objects_softmax_cast_fp16)[name = string("out_objects_cast_fp16")];
+            string out_objects_cast_fp16_to_fp32_dtype_0 = const()[name = string("out_objects_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp32, [1, ?, 1025]> log_probs = cast(dtype = out_objects_cast_fp16_to_fp32_dtype_0, x = out_objects_cast_fp16)[name = string("cast_0")];
+        } -> (log_probs);
+}

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a81ba8ca75a2a6bed883f6dc219d52bfe3ac9cec5929230f3ea577d47bed51e5
+size 527170

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5a32996f7cdfb5cb2d252122710546e08cf98c7b1743878e9be8e82f2a31ec2
+size 243

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9981193e84379ddd5c84a17ed58e95fa4185e49cdc6c075834a3279fefc9350e
+size 496

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/metadata.json

@@ -0,0 +1,117 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC Mel+Encoder (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 188)",
+        "shortDescription" : "",
+        "shape" : "[1, 1024, 188]",
+        "name" : "encoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "encoder_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int32, Palettized (4 bits), UInt4)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Stack" : 1,
+      "Ios18.conv" : 79,
+      "Ios18.mul" : 107,
+      "Ios18.constexprLutToDense" : 298,
+      "Ios18.cast" : 12,
+      "Ios18.layerNorm" : 120,
+      "Ios18.log" : 1,
+      "Ios18.equal" : 1,
+      "Ios18.floorDiv" : 4,
+      "Ios16.reduceSum" : 4,
+      "Ios18.logicalNot" : 3,
+      "Ios18.reshape" : 147,
+      "Pad" : 49,
+      "Ios18.concat" : 1,
+      "Ios18.add" : 180,
+      "Ios18.realDiv" : 3,
+      "Ios18.greaterEqual" : 1,
+      "Ios18.notEqual" : 1,
+      "Ios18.relu" : 3,
+      "Ios18.sub" : 7,
+      "Ios18.matmul" : 73,
+      "Ios18.silu" : 72,
+      "Ios18.expandDims" : 25,
+      "Ios18.linear" : 193,
+      "Ios18.sigmoid" : 24,
+      "Ios18.sliceByIndex" : 51,
+      "Ios18.transpose" : 172,
+      "Split" : 24,
+      "Ios18.sqrt" : 1,
+      "Ios18.softmax" : 24,
+      "Ios18.pow" : 2,
+      "Select" : 78,
+      "Ios18.logicalAnd" : 2,
+      "Ios18.less" : 6,
+      "Tile" : 9
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 240000)",
+        "shortDescription" : "",
+        "shape" : "[1, 240000]",
+        "name" : "audio_signal",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "audio_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_mel_encoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/4bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:971bdc0f0bfc738b0d9e7fe6aca1d30fee5c2a55c15a8f8955b07767db1a76d0
+size 298251072

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7000c84a4ca4b88f7c286df332b8a5b6cf98d6f5472352eea08470f38f46cdc1
+size 243

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bb8115c6d918526b13db13e228e22dd8010269a545d922d83d6a10ab0dedc42
+size 463

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_decoder.mlmodelc/metadata.json

@@ -0,0 +1,70 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC decoder head (encoder -> log_probs)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "log_probs",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Palettized (6 bits), UInt6)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.transpose" : 1,
+      "Ios18.constexprLutToDense" : 1,
+      "Ios18.cast" : 2,
+      "Ios18.conv" : 1,
+      "Ios18.softmax" : 1,
+      "Ios18.log" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 1024 × 1...188",
+        "shapeRange" : "[[1, 1], [1024, 1024], [1, 188]]",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 1)",
+        "type" : "MultiArray",
+        "shape" : "[1, 1024, 1]",
+        "name" : "encoder",
+        "shortDescription" : ""
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_decoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_decoder.mlmodelc/model.mil

@@ -0,0 +1,24 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
+{
+    func main<ios18>(tensor<fp32, [1, 1024, ?]> encoder) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"encoder", [1, 1024, 1]}}), ("RangeDims", {{"encoder", [[1, 1], [1024, 1024], [1, 188]]}})))] {
+            int32 var_4 = const()[name = string("op_4"), val = int32(-1)];
+            string var_18_pad_type_0 = const()[name = string("op_18_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> var_18_strides_0 = const()[name = string("op_18_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_18_pad_0 = const()[name = string("op_18_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_18_dilations_0 = const()[name = string("op_18_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 var_18_groups_0 = const()[name = string("op_18_groups_0"), val = int32(1)];
+            string encoder_to_fp16_dtype_0 = const()[name = string("encoder_to_fp16_dtype_0"), val = string("fp16")];
+            tensor<fp16, [1025, 1024, 1]> module_decoder_layers_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1025, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(787328))))[name = string("module_decoder_layers_0_weight_to_fp16_palettized")];
+            tensor<fp16, [1025]> module_decoder_layers_0_bias_to_fp16 = const()[name = string("module_decoder_layers_0_bias_to_fp16"), val = tensor<fp16, [1025]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(787520)))];
+            tensor<fp16, [1, 1024, ?]> encoder_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = string("cast_1")];
+            tensor<fp16, [1, 1025, ?]> var_18_cast_fp16 = conv(bias = module_decoder_layers_0_bias_to_fp16, dilations = var_18_dilations_0, groups = var_18_groups_0, pad = var_18_pad_0, pad_type = var_18_pad_type_0, strides = var_18_strides_0, weight = module_decoder_layers_0_weight_to_fp16_palettized, x = encoder_to_fp16)[name = string("op_18_cast_fp16")];
+            tensor<int32, [3]> input_perm_0 = const()[name = string("input_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, ?, 1025]> input_cast_fp16 = transpose(perm = input_perm_0, x = var_18_cast_fp16)[name = string("transpose_0")];
+            tensor<fp16, [1, ?, 1025]> out_objects_softmax_cast_fp16 = softmax(axis = var_4, x = input_cast_fp16)[name = string("out_objects_softmax_cast_fp16")];
+            fp32 out_objects_epsilon_0 = const()[name = string("out_objects_epsilon_0"), val = fp32(0x1p-149)];
+            tensor<fp16, [1, ?, 1025]> out_objects_cast_fp16 = log(epsilon = out_objects_epsilon_0, x = out_objects_softmax_cast_fp16)[name = string("out_objects_cast_fp16")];
+            string out_objects_cast_fp16_to_fp32_dtype_0 = const()[name = string("out_objects_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp32, [1, ?, 1025]> log_probs = cast(dtype = out_objects_cast_fp16_to_fp32_dtype_0, x = out_objects_cast_fp16)[name = string("cast_0")];
+        } -> (log_probs);
+}

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d0a073770791d8ff2214fc89dcb29e6e42a11c0c253f0c3c0a23d3e79ad69ec
+size 789634

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:124ea4cda0d5d6210203fd724f81e5284e78fea6a9517013f25d54b046a393d6
+size 243

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85e9f93f4605ea3f79760b8005fc26ba2ff209c2f1865af09cfb76805c1c8938
+size 496

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/metadata.json

@@ -0,0 +1,117 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC Mel+Encoder (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 188)",
+        "shortDescription" : "",
+        "shape" : "[1, 1024, 188]",
+        "name" : "encoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "encoder_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int32, Palettized (6 bits), UInt6)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Stack" : 1,
+      "Ios18.conv" : 79,
+      "Ios18.mul" : 107,
+      "Ios18.constexprLutToDense" : 298,
+      "Ios18.cast" : 12,
+      "Ios18.layerNorm" : 120,
+      "Ios18.log" : 1,
+      "Ios18.equal" : 1,
+      "Ios18.floorDiv" : 4,
+      "Ios16.reduceSum" : 4,
+      "Ios18.logicalNot" : 3,
+      "Ios18.reshape" : 147,
+      "Pad" : 49,
+      "Ios18.concat" : 1,
+      "Ios18.add" : 180,
+      "Ios18.realDiv" : 3,
+      "Ios18.greaterEqual" : 1,
+      "Ios18.notEqual" : 1,
+      "Ios18.relu" : 3,
+      "Ios18.sub" : 7,
+      "Ios18.matmul" : 73,
+      "Ios18.silu" : 72,
+      "Ios18.expandDims" : 25,
+      "Ios18.linear" : 193,
+      "Ios18.sigmoid" : 24,
+      "Ios18.sliceByIndex" : 51,
+      "Ios18.transpose" : 172,
+      "Split" : 24,
+      "Ios18.sqrt" : 1,
+      "Ios18.softmax" : 24,
+      "Ios18.pow" : 2,
+      "Select" : 78,
+      "Ios18.logicalAnd" : 2,
+      "Ios18.less" : 6,
+      "Tile" : 9
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 240000)",
+        "shortDescription" : "",
+        "shape" : "[1, 240000]",
+        "name" : "audio_signal",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "audio_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_mel_encoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/6bit_palettize/parakeet_ctc_mel_encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac7e95814789c229224f8971023134c2175a2634685014a25bff941539870953
+size 446123712

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8457b1bdabeec9901e2fd226bcd0f8eec18090f6268240e0fef2405f3b068ccb
+size 243

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:279d53e635ca4b9da206fbaad927a63def64702e23913df3c0b02e8dcf4edf15
+size 463

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_decoder.mlmodelc/metadata.json

@@ -0,0 +1,70 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC decoder head (encoder -> log_probs)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "log_probs",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int4)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.transpose" : 1,
+      "Ios18.log" : 1,
+      "Ios18.cast" : 2,
+      "Ios18.conv" : 1,
+      "Ios18.softmax" : 1,
+      "Ios18.constexprBlockwiseShiftScale" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 1024 × 1...188",
+        "shapeRange" : "[[1, 1], [1024, 1024], [1, 188]]",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 1)",
+        "type" : "MultiArray",
+        "shape" : "[1, 1024, 1]",
+        "name" : "encoder",
+        "shortDescription" : ""
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_decoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_decoder.mlmodelc/model.mil

@@ -0,0 +1,24 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
+{
+    func main<ios18>(tensor<fp32, [1, 1024, ?]> encoder) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"encoder", [1, 1024, 1]}}), ("RangeDims", {{"encoder", [[1, 1], [1024, 1024], [1, 188]]}})))] {
+            int32 var_4 = const()[name = string("op_4"), val = int32(-1)];
+            string var_18_pad_type_0 = const()[name = string("op_18_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> var_18_strides_0 = const()[name = string("op_18_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_18_pad_0 = const()[name = string("op_18_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_18_dilations_0 = const()[name = string("op_18_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 var_18_groups_0 = const()[name = string("op_18_groups_0"), val = int32(1)];
+            string encoder_to_fp16_dtype_0 = const()[name = string("encoder_to_fp16_dtype_0"), val = string("fp16")];
+            tensor<fp16, [1025, 1024, 1]> module_decoder_layers_0_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1025, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [1025, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(524928))))[name = string("module_decoder_layers_0_weight_to_fp16_quantized")];
+            tensor<fp16, [1025]> module_decoder_layers_0_bias_to_fp16 = const()[name = string("module_decoder_layers_0_bias_to_fp16"), val = tensor<fp16, [1025]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(590592)))];
+            tensor<fp16, [1, 1024, ?]> encoder_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = string("cast_1")];
+            tensor<fp16, [1, 1025, ?]> var_18_cast_fp16 = conv(bias = module_decoder_layers_0_bias_to_fp16, dilations = var_18_dilations_0, groups = var_18_groups_0, pad = var_18_pad_0, pad_type = var_18_pad_type_0, strides = var_18_strides_0, weight = module_decoder_layers_0_weight_to_fp16_quantized, x = encoder_to_fp16)[name = string("op_18_cast_fp16")];
+            tensor<int32, [3]> input_perm_0 = const()[name = string("input_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, ?, 1025]> input_cast_fp16 = transpose(perm = input_perm_0, x = var_18_cast_fp16)[name = string("transpose_0")];
+            tensor<fp16, [1, ?, 1025]> out_objects_softmax_cast_fp16 = softmax(axis = var_4, x = input_cast_fp16)[name = string("out_objects_softmax_cast_fp16")];
+            fp32 out_objects_epsilon_0 = const()[name = string("out_objects_epsilon_0"), val = fp32(0x1p-149)];
+            tensor<fp16, [1, ?, 1025]> out_objects_cast_fp16 = log(epsilon = out_objects_epsilon_0, x = out_objects_softmax_cast_fp16)[name = string("out_objects_cast_fp16")];
+            string out_objects_cast_fp16_to_fp32_dtype_0 = const()[name = string("out_objects_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp32, [1, ?, 1025]> log_probs = cast(dtype = out_objects_cast_fp16_to_fp32_dtype_0, x = out_objects_cast_fp16)[name = string("cast_0")];
+        } -> (log_probs);
+}

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0e1e88ab1787121c69b1f505d5789f627d95336cd0b0a69079fb5ddec59950b
+size 592706

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_mel_encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3e1ff6f1f3c5a8ccac117bd945c0bf771bf7e298df6c9a5363ddbf7d5ee8cd0
+size 243

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_mel_encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e49899cd89d6a7d8137a46cd91e7a9b266c26649e3fc0ff6a45f2d352e4bc4f6
+size 496

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_mel_encoder.mlmodelc/metadata.json

@@ -0,0 +1,117 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC Mel+Encoder (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 188)",
+        "shortDescription" : "",
+        "shape" : "[1, 1024, 188]",
+        "name" : "encoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "encoder_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int32, Int4)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.less" : 6,
+      "Stack" : 1,
+      "Ios18.conv" : 79,
+      "Ios18.mul" : 107,
+      "Ios18.cast" : 12,
+      "Ios18.layerNorm" : 120,
+      "Ios18.log" : 1,
+      "Ios18.equal" : 1,
+      "Ios18.floorDiv" : 4,
+      "Ios16.reduceSum" : 4,
+      "Ios18.logicalNot" : 3,
+      "Ios18.reshape" : 147,
+      "Pad" : 49,
+      "Ios18.concat" : 1,
+      "Ios18.add" : 180,
+      "Ios18.realDiv" : 3,
+      "Ios18.greaterEqual" : 1,
+      "Ios18.notEqual" : 1,
+      "Ios18.relu" : 3,
+      "Ios18.sub" : 7,
+      "Ios18.matmul" : 73,
+      "Ios18.silu" : 72,
+      "Ios18.expandDims" : 25,
+      "Ios18.linear" : 193,
+      "Ios18.sigmoid" : 24,
+      "Ios18.sliceByIndex" : 51,
+      "Ios18.transpose" : 172,
+      "Split" : 24,
+      "Ios18.sqrt" : 1,
+      "Ios18.softmax" : 24,
+      "Ios18.pow" : 2,
+      "Select" : 78,
+      "Ios18.constexprBlockwiseShiftScale" : 267,
+      "Ios18.logicalAnd" : 2,
+      "Tile" : 9
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 240000)",
+        "shortDescription" : "",
+        "shape" : "[1, 240000]",
+        "name" : "audio_signal",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "audio_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_mel_encoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/int4_linear/parakeet_ctc_mel_encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c0cdf923b883bc26edbea473afdfddca4c827d5633ff185221cde0ee0ef8f76
+size 335597952

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc3479032d2ad09d5f2e4d4adf9dd8a22d723cf4d0e50bd1cfec2d6e7bb9249b
+size 243

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d3963850ea80c07531b38128cd51f06b6dbc4d138a9c15f9834e72977c9b2fa
+size 463

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_decoder.mlmodelc/metadata.json

@@ -0,0 +1,70 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC decoder head (encoder -> log_probs)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32)",
+        "shortDescription" : "",
+        "shape" : "[]",
+        "name" : "log_probs",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int8)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.transpose" : 1,
+      "Ios18.log" : 1,
+      "Ios18.cast" : 2,
+      "Ios18.conv" : 1,
+      "Ios18.softmax" : 1,
+      "Ios18.constexprBlockwiseShiftScale" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "dataType" : "Float32",
+        "hasShapeFlexibility" : "1",
+        "isOptional" : "0",
+        "shapeFlexibility" : "1 × 1024 × 1...188",
+        "shapeRange" : "[[1, 1], [1024, 1024], [1, 188]]",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 1)",
+        "type" : "MultiArray",
+        "shape" : "[1, 1024, 1]",
+        "name" : "encoder",
+        "shortDescription" : ""
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_decoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_decoder.mlmodelc/model.mil

@@ -0,0 +1,24 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
+{
+    func main<ios18>(tensor<fp32, [1, 1024, ?]> encoder) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"encoder", [1, 1024, 1]}}), ("RangeDims", {{"encoder", [[1, 1], [1024, 1024], [1, 188]]}})))] {
+            int32 var_4 = const()[name = string("op_4"), val = int32(-1)];
+            string var_18_pad_type_0 = const()[name = string("op_18_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> var_18_strides_0 = const()[name = string("op_18_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_18_pad_0 = const()[name = string("op_18_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_18_dilations_0 = const()[name = string("op_18_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 var_18_groups_0 = const()[name = string("op_18_groups_0"), val = int32(1)];
+            string encoder_to_fp16_dtype_0 = const()[name = string("encoder_to_fp16_dtype_0"), val = string("fp16")];
+            tensor<fp16, [1025, 1024, 1]> module_decoder_layers_0_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1025, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), offset = tensor<int8, [1025, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1051904))), scale = tensor<fp16, [1025, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1049728))))[name = string("module_decoder_layers_0_weight_to_fp16_quantized")];
+            tensor<fp16, [1025]> module_decoder_layers_0_bias_to_fp16 = const()[name = string("module_decoder_layers_0_bias_to_fp16"), val = tensor<fp16, [1025]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1053056)))];
+            tensor<fp16, [1, 1024, ?]> encoder_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = string("cast_1")];
+            tensor<fp16, [1, 1025, ?]> var_18_cast_fp16 = conv(bias = module_decoder_layers_0_bias_to_fp16, dilations = var_18_dilations_0, groups = var_18_groups_0, pad = var_18_pad_0, pad_type = var_18_pad_type_0, strides = var_18_strides_0, weight = module_decoder_layers_0_weight_to_fp16_quantized, x = encoder_to_fp16)[name = string("op_18_cast_fp16")];
+            tensor<int32, [3]> input_perm_0 = const()[name = string("input_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, ?, 1025]> input_cast_fp16 = transpose(perm = input_perm_0, x = var_18_cast_fp16)[name = string("transpose_0")];
+            tensor<fp16, [1, ?, 1025]> out_objects_softmax_cast_fp16 = softmax(axis = var_4, x = input_cast_fp16)[name = string("out_objects_softmax_cast_fp16")];
+            fp32 out_objects_epsilon_0 = const()[name = string("out_objects_epsilon_0"), val = fp32(0x1p-149)];
+            tensor<fp16, [1, ?, 1025]> out_objects_cast_fp16 = log(epsilon = out_objects_epsilon_0, x = out_objects_softmax_cast_fp16)[name = string("out_objects_cast_fp16")];
+            string out_objects_cast_fp16_to_fp32_dtype_0 = const()[name = string("out_objects_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp32, [1, ?, 1025]> log_probs = cast(dtype = out_objects_cast_fp16_to_fp32_dtype_0, x = out_objects_cast_fp16)[name = string("cast_0")];
+        } -> (log_probs);
+}

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a51b1b7459ddae4323b821ec75cbecc339867372f60602bb1e32d546a0817bf6
+size 1055170

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_mel_encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b3e5c4a76594e4cb6948b59ba533672f57daa7670531ae91a9cc8b95ecea1db
+size 243

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_mel_encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84828b5ad1820337049ab436c70bf66017b30153e26774dd986866e769b5a5b6
+size 496

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_mel_encoder.mlmodelc/metadata.json

@@ -0,0 +1,117 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet CTC Mel+Encoder (15.0s window)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 188)",
+        "shortDescription" : "",
+        "shape" : "[1, 1024, 188]",
+        "name" : "encoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "encoder_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Mixed (Float16, Int32, Int8)",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.constexprBlockwiseShiftScale" : 298,
+      "Stack" : 1,
+      "Ios18.conv" : 79,
+      "Ios18.mul" : 107,
+      "Ios18.cast" : 12,
+      "Ios18.layerNorm" : 120,
+      "Ios18.log" : 1,
+      "Ios18.equal" : 1,
+      "Ios18.floorDiv" : 4,
+      "Ios16.reduceSum" : 4,
+      "Ios18.logicalNot" : 3,
+      "Ios18.reshape" : 147,
+      "Pad" : 49,
+      "Ios18.concat" : 1,
+      "Ios18.add" : 180,
+      "Ios18.realDiv" : 3,
+      "Ios18.relu" : 3,
+      "Ios18.greaterEqual" : 1,
+      "Ios18.notEqual" : 1,
+      "Ios18.sub" : 7,
+      "Ios18.matmul" : 73,
+      "Ios18.silu" : 72,
+      "Ios18.expandDims" : 25,
+      "Ios18.linear" : 193,
+      "Ios18.sigmoid" : 24,
+      "Ios18.sliceByIndex" : 51,
+      "Ios18.transpose" : 172,
+      "Split" : 24,
+      "Ios18.sqrt" : 1,
+      "Ios18.softmax" : 24,
+      "Ios18.pow" : 2,
+      "Select" : 78,
+      "Ios18.logicalAnd" : 2,
+      "Ios18.less" : 6,
+      "Tile" : 9
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 240000)",
+        "shortDescription" : "",
+        "shape" : "[1, 240000]",
+        "name" : "audio_signal",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "audio_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-27",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_ctc_mel_encoder",
+    "method" : "predict"
+  }
+]

compiled/parakeet_ctc_coreml_quantized/int8_linear/parakeet_ctc_mel_encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d93fd59e9f558fa8127254aafdf5c68927ddb3b42ad7d4cdc7a3e330faea9f8
+size 595321920