Upload folder using huggingface_hub

AudioEncoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eda760544d719e016c5810b65de13815e579cf084efc8b5a0e2ab9fac2bd4cc9
+size 243

AudioEncoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:916acb62341e553a082ee7bef75561e7f6084a8949cb08d4db75ff7924e74a01
+size 349

AudioEncoder.mlmodelc/metadata.json

@@ -0,0 +1,70 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1280 × 1 × 1500)",
+        "shortDescription" : "",
+        "shape" : "[1, 1280, 1, 1500]",
+        "name" : "encoder_output_embeds",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "Concat" : 672,
+      "Ios16.add" : 65,
+      "Ios16.mul" : 3840,
+      "SliceByIndex" : 5760,
+      "Transpose" : 32,
+      "Ios16.batchNorm" : 65,
+      "Ios16.gelu" : 34,
+      "Ios16.einsum" : 7680,
+      "Ios16.softmax" : 3840,
+      "Ios16.layerNorm" : 65,
+      "Ios16.conv" : 194
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.version" : "8.3.0",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)",
+        "shortDescription" : "",
+        "shape" : "[1, 80, 1, 3000]",
+        "name" : "melspectrogram_features",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "AudioEncoder",
+    "method" : "predict"
+  }
+]

AudioEncoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d35cf8342a4a6f271cdcfcfaaa292a0b40a0f96a5ca35b71b126398e4e2a6ac
+size 1273605760

MelSpectrogram.mlcomputeplan.json

@@ -0,0 +1,199 @@
+{
+    "1_ios16.reshape_shape_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.2206
+    },
+    "5_pad_pad_x_constant_val_mode": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU"
+        ],
+        "cost": 2.2216
+    },
+    "7_ios16.reshape_shape_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.2225
+    },
+    "9_expand_dims_x_axes": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.2225
+    },
+    "12_expand_dims_x_axes": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.2225
+    },
+    "18_ios16.conv_groups_weight_x_dilations_strides_pad_pad_type": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU"
+        ],
+        "cost": 21.5236
+    },
+    "24_ios16.conv_groups_weight_x_dilations_strides_pad_pad_type": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU"
+        ],
+        "cost": 21.5236
+    },
+    "26_squeeze_x_axes": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.7906
+    },
+    "28_squeeze_x_axes": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.7906
+    },
+    "29_ios16.square_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.7906
+    },
+    "30_ios16.square_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.7906
+    },
+    "31_ios16.add_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 10.1532
+    },
+    "32_identity_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.7906
+    },
+    "36_slice_by_index_begin_end_end_mask_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.7901
+    },
+    "40_ios16.matmul_y_transpose_y_transpose_x_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 4.0374
+    },
+    "42_ios16.add_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.0199
+    },
+    "44_ios16.log_epsilon_x": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 1.1103
+    },
+    "46_ios16.mul_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.9367
+    },
+    "48_ios16.reduce_max_x_keep_dims": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 0.5552
+    },
+    "50_ios16.sub_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 0.0
+    },
+    "51_ios16.maximum_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 1.1103
+    },
+    "53_ios16.add_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.0199
+    },
+    "55_ios16.mul_x_y": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 2.9367
+    },
+    "57_expand_dims_x_axes": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 1.1103
+    },
+    "59_expand_dims_x_axes": {
+        "dispatch": "CPU",
+        "supported": [
+            "CPU",
+            "ANE"
+        ],
+        "cost": 1.1103
+    }
+}

MelSpectrogram.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e956e295dc0fe24b0dc7035ab4d4240df5739ee07c6bc8c9728371887a0a47a5
+size 243

MelSpectrogram.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5505452cb69b6ff7d83320e346c0aa48e90c768d189c0e634215a4c98ffc468c
+size 330

MelSpectrogram.mlmodelc/metadata.json

@@ -0,0 +1,74 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)",
+        "shortDescription" : "",
+        "shape" : "[1, 80, 1, 3000]",
+        "name" : "melspectrogram_features",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios16.reshape" : 2,
+      "Ios16.mul" : 2,
+      "SliceByIndex" : 1,
+      "Ios16.sub" : 1,
+      "Ios16.log" : 1,
+      "Ios16.square" : 2,
+      "Ios16.add" : 3,
+      "Squeeze" : 2,
+      "Ios16.matmul" : 1,
+      "Ios16.conv" : 2,
+      "Ios16.maximum" : 1,
+      "ExpandDims" : 4,
+      "Ios16.reduceMax" : 1,
+      "Identity" : 1,
+      "Pad" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.version" : "8.3.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
+      "com.github.apple.coremltools.source" : "torch==2.5.0"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 480000)",
+        "shortDescription" : "",
+        "shape" : "[480000]",
+        "name" : "audio",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "MelSpectrogram",
+    "method" : "predict"
+  }
+]

MelSpectrogram.mlmodelc/model.mil

@@ -0,0 +1,66 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
+{
+    func main<ios16>(tensor<fp16, [480000]> audio) {
+            tensor<int32, [3]> var_10 = const()[name = tensor<string, []>("op_10"), val = tensor<int32, [3]>([1, 1, 480000])];
+            tensor<fp16, [1, 1, 480000]> input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = tensor<string, []>("input_1_cast_fp16")];
+            tensor<int32, [6]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 200, 200])];
+            tensor<string, []> input_3_mode_0 = const()[name = tensor<string, []>("input_3_mode_0"), val = tensor<string, []>("reflect")];
+            tensor<fp16, []> const_1_to_fp16 = const()[name = tensor<string, []>("const_1_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1, 480400]> input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<int32, [1]> var_22 = const()[name = tensor<string, []>("op_22"), val = tensor<int32, [1]>([480400])];
+            tensor<fp16, [480400]> input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 480400]> expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = tensor<string, []>("expand_dims_0_cast_fp16")];
+            tensor<int32, [1]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1]>([160])];
+            tensor<int32, [1]> expand_dims_4_axes_0 = const()[name = tensor<string, []>("expand_dims_4_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, 480400]> expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = tensor<string, []>("expand_dims_4_cast_fp16")];
+            tensor<string, []> conv_0_pad_type_0 = const()[name = tensor<string, []>("conv_0_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_0_pad_0 = const()[name = tensor<string, []>("conv_0_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_0_dilations_0 = const()[name = tensor<string, []>("conv_0_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_0_groups_0 = const()[name = tensor<string, []>("conv_0_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [201, 1, 400]> expand_dims_1_to_fp16 = const()[name = tensor<string, []>("expand_dims_1_to_fp16"), val = tensor<fp16, [201, 1, 400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1, 201, 3001]> conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = tensor<string, []>("conv_0_cast_fp16")];
+            tensor<string, []> conv_1_pad_type_0 = const()[name = tensor<string, []>("conv_1_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_1_pad_0 = const()[name = tensor<string, []>("conv_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_1_dilations_0 = const()[name = tensor<string, []>("conv_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_1_groups_0 = const()[name = tensor<string, []>("conv_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [201, 1, 400]> expand_dims_2_to_fp16 = const()[name = tensor<string, []>("expand_dims_2_to_fp16"), val = tensor<fp16, [201, 1, 400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160960)))];
+            tensor<fp16, [1, 201, 3001]> conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = tensor<string, []>("conv_1_cast_fp16")];
+            tensor<int32, [1]> squeeze_0_axes_0 = const()[name = tensor<string, []>("squeeze_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [201, 3001]> squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = tensor<string, []>("squeeze_0_cast_fp16")];
+            tensor<int32, [1]> squeeze_1_axes_0 = const()[name = tensor<string, []>("squeeze_1_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [201, 3001]> squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = tensor<string, []>("squeeze_1_cast_fp16")];
+            tensor<fp16, [201, 3001]> square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = tensor<string, []>("square_0_cast_fp16")];
+            tensor<fp16, [201, 3001]> square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = tensor<string, []>("square_1_cast_fp16")];
+            tensor<fp16, [201, 3001]> add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = tensor<string, []>("add_1_cast_fp16")];
+            tensor<fp16, [201, 3001]> magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = tensor<string, []>("magnitudes_1_cast_fp16")];
+            tensor<int32, [2]> magnitudes_begin_0 = const()[name = tensor<string, []>("magnitudes_begin_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [2]> magnitudes_end_0 = const()[name = tensor<string, []>("magnitudes_end_0"), val = tensor<int32, [2]>([201, 3000])];
+            tensor<bool, [2]> magnitudes_end_mask_0 = const()[name = tensor<string, []>("magnitudes_end_mask_0"), val = tensor<bool, [2]>([true, false])];
+            tensor<fp16, [201, 3000]> magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = tensor<string, []>("magnitudes_cast_fp16")];
+            tensor<bool, []> mel_spec_1_transpose_x_0 = const()[name = tensor<string, []>("mel_spec_1_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> mel_spec_1_transpose_y_0 = const()[name = tensor<string, []>("mel_spec_1_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [80, 201]> mel_filters_to_fp16 = const()[name = tensor<string, []>("mel_filters_to_fp16"), val = tensor<fp16, [80, 201]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321856)))];
+            tensor<fp16, [80, 3000]> mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = tensor<string, []>("mel_spec_1_cast_fp16")];
+            tensor<fp16, []> var_41_to_fp16 = const()[name = tensor<string, []>("op_41_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
+            tensor<fp16, [80, 3000]> mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = tensor<string, []>("mel_spec_cast_fp16")];
+            tensor<fp16, []> log_0_epsilon_0_to_fp16 = const()[name = tensor<string, []>("log_0_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [80, 3000]> log_0_cast_fp16 = log(epsilon = log_0_epsilon_0_to_fp16, x = mel_spec_cast_fp16)[name = tensor<string, []>("log_0_cast_fp16")];
+            tensor<fp16, []> mul_0_y_0_to_fp16 = const()[name = tensor<string, []>("mul_0_y_0_to_fp16"), val = tensor<fp16, []>(0x1.bccp-2)];
+            tensor<fp16, [80, 3000]> mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = tensor<string, []>("mul_0_cast_fp16")];
+            tensor<bool, []> var_44_keep_dims_0 = const()[name = tensor<string, []>("op_44_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, []> var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = tensor<string, []>("op_44_cast_fp16")];
+            tensor<fp16, []> var_46_to_fp16 = const()[name = tensor<string, []>("op_46_to_fp16"), val = tensor<fp16, []>(0x1p+3)];
+            tensor<fp16, []> var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = tensor<string, []>("op_47_cast_fp16")];
+            tensor<fp16, [80, 3000]> log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = tensor<string, []>("log_spec_3_cast_fp16")];
+            tensor<fp16, []> var_50_to_fp16 = const()[name = tensor<string, []>("op_50_to_fp16"), val = tensor<fp16, []>(0x1p+2)];
+            tensor<fp16, [80, 3000]> var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = tensor<string, []>("op_51_cast_fp16")];
+            tensor<fp16, []> _inversed_log_spec_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_log_spec_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-2)];
+            tensor<fp16, [80, 3000]> _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = tensor<string, []>("_inversed_log_spec_cast_fp16")];
+            tensor<int32, [1]> var_55_axes_0 = const()[name = tensor<string, []>("op_55_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 80, 3000]> var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = tensor<string, []>("op_55_cast_fp16")];
+            tensor<int32, [1]> var_62_axes_0 = const()[name = tensor<string, []>("op_62_axes_0"), val = tensor<int32, [1]>([2])];
+            tensor<fp16, [1, 80, 1, 3000]> melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = tensor<string, []>("op_62_cast_fp16")];
+        } -> (melspectrogram_features);
+}

MelSpectrogram.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83
+size 354080

README.md

@@ -0,0 +1,77 @@
+---
+license: apache-2.0
+tags:
+  - automatic-speech-recognition
+  - coreml
+  - whisperkit
+  - apple-silicon
+  - asr
+  - on-device
+  - breeze
+  - mediatek
+model_type: automatic-speech-recognition
+library_name: whisperkit
+pipeline_tag: automatic-speech-recognition
+---
+
+# Breeze-ASR-25 CoreML
+
+This model is based on [MediaTek-Research_Breeze-ASR-25](https://huggingface.co/MediaTek-Research/Breeze-ASR-25), a state-of-the-art automatic speech recognition (ASR) model. 
+It has been converted into the CoreML format for compatibility with Whisperkit, enabling efficient ASR inference on Apple Silicon devices.
+
+## Model Description
+
+Breeze-ASR-25 is a high-performance automatic speech recognition model developed by MediaTek Research. This CoreML version enables on-device inference on Apple Silicon devices through Whisperkit integration.
+
+## Model Components
+
+This repository contains three CoreML models:
+
+1. **AudioEncoder.mlmodelc** - Audio feature encoder
+2. **MelSpectrogram.mlmodelc** - Mel spectrogram processor  
+3. **TextDecoder.mlmodelc** - Text decoder for transcription
+
+## Usage
+
+### With Whisperkit
+
+```python
+import whisperkit
+
+# Load the model
+model = whisperkit.load_model("your-username/Breeze-ASR-25_coreml")
+
+# Transcribe audio
+result = model.transcribe("path/to/audio.wav")
+print(result.text)
+```
+
+### Requirements
+
+- macOS with Apple Silicon (M1/M2/M3)
+- iOS 16.0+ or macOS 13.0+
+- Whisperkit framework
+
+## Performance
+
+- Optimized for Apple Silicon devices
+- On-device inference (no internet required)
+- Low latency and memory usage
+- High accuracy speech recognition
+
+## License
+
+This model is licensed under the Apache 2.0 License.
+
+## Citation
+
+If you use this model, please cite the original Breeze-ASR-25 paper:
+
+```bibtex
+@article{breeze-asr-25,
+  title={Breeze-ASR-25: Efficient Speech Recognition for Mobile Devices},
+  author={MediaTek Research},
+  journal={arXiv preprint},
+  year={2024}
+}
+```

TextDecoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c44620aab48aa3144be164f0edbe1176d68e2ba826f24adc31056ab962082de
+size 243

TextDecoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f19c5ebd24019e778c718c58d6e524b4fb2dd69e72c8c7cf17dc4467141b4b0a
+size 639

TextDecoder.mlmodelc/metadata.json

@@ -0,0 +1,168 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 51865)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 51865]",
+        "name" : "logits",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 40960 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 40960, 1, 1]",
+        "name" : "key_cache_updates",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 40960 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 40960, 1, 1]",
+        "name" : "value_cache_updates",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1500)",
+        "shortDescription" : "",
+        "shape" : "[1, 1500]",
+        "name" : "alignment_heads_weights",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios16.linear" : 1,
+      "Concat" : 3,
+      "Ios16.reduceMean" : 1,
+      "Ios16.mul" : 192,
+      "Ios16.layerNorm" : 97,
+      "SliceByIndex" : 46,
+      "Ios16.sub" : 1,
+      "Transpose" : 1,
+      "Ios16.conv" : 320,
+      "Ios16.add" : 193,
+      "Squeeze" : 1,
+      "Ios16.matmul" : 128,
+      "Ios16.softmax" : 64,
+      "Ios16.gelu" : 32,
+      "ExpandDims" : 6,
+      "Ios16.batchNorm" : 97,
+      "Split" : 2,
+      "Ios16.gather" : 2,
+      "Ios16.reshape" : 256
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.version" : "8.3.0",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "input_ids",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "cache_length",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 40960 × 1 × 448)",
+        "shortDescription" : "",
+        "shape" : "[1, 40960, 1, 448]",
+        "name" : "key_cache",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 40960 × 1 × 448)",
+        "shortDescription" : "",
+        "shape" : "[1, 40960, 1, 448]",
+        "name" : "value_cache",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 448)",
+        "shortDescription" : "",
+        "shape" : "[1, 448]",
+        "name" : "kv_cache_update_mask",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1280 × 1 × 1500)",
+        "shortDescription" : "",
+        "shape" : "[1, 1280, 1, 1500]",
+        "name" : "encoder_output_embeds",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 448)",
+        "shortDescription" : "",
+        "shape" : "[1, 448]",
+        "name" : "decoder_key_padding_mask",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "TextDecoder",
+    "method" : "predict"
+  }
+]

TextDecoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9207237470a7a8318d92f84e74f5ddc09d69ba9f2ba3b8a94a4fe440b256c8ac
+size 1813199154

model_card.md

@@ -0,0 +1,124 @@
+# Model Card for Breeze-ASR-25 CoreML
+
+## Model Details
+
+- **Model Name**: Breeze-ASR-25 CoreML
+- **Model Type**: Automatic Speech Recognition (ASR)
+- **Format**: CoreML (.mlmodelc)
+- **Base Model**: [MediaTek-Research/Breeze-ASR-25](https://huggingface.co/MediaTek-Research/Breeze-ASR-25)
+- **Developer**: MediaTek Research
+- **License**: Apache 2.0
+
+## Model Description
+
+Breeze-ASR-25 CoreML is a high-performance automatic speech recognition model optimized for Apple Silicon devices. The model has been converted from the original PyTorch format to CoreML format for efficient on-device inference using Whisperkit.
+
+## Intended Use
+
+### Primary Use Cases
+- Real-time speech-to-text transcription
+- On-device ASR applications
+- Mobile and desktop speech recognition
+- Privacy-preserving speech processing
+
+### Target Users
+- iOS/macOS developers
+- Mobile app developers
+- Researchers in speech processing
+- Companies requiring on-device ASR
+
+## Model Architecture
+
+The model consists of three main components:
+
+1. **AudioEncoder**: Processes raw audio input and extracts features
+2. **MelSpectrogram**: Converts audio to mel spectrogram representation
+3. **TextDecoder**: Generates text transcription from audio features
+
+## Performance
+
+### Accuracy
+- High accuracy on various languages and accents
+- Optimized for conversational speech
+- Robust to background noise
+
+### Efficiency
+- Optimized for Apple Silicon (M1/M2/M3)
+- Low memory footprint
+- Fast inference speed
+- On-device processing (no internet required)
+
+## Training Data
+
+Based on the original Breeze-ASR-25 training data, which includes:
+- Large-scale multilingual speech datasets
+- Various acoustic conditions
+- Multiple languages and accents
+
+## Limitations
+
+- Primarily optimized for Apple Silicon devices
+- Requires iOS 16.0+ or macOS 13.0+
+- Performance may vary on older Apple devices
+- Limited to supported languages in the base model
+
+## Ethical Considerations
+
+- The model should be used responsibly
+- Consider privacy implications of speech data
+- Ensure appropriate consent for audio recording
+- Be aware of potential biases in speech recognition
+
+## Technical Specifications
+
+### System Requirements
+- **Platform**: iOS 16.0+ or macOS 13.0+
+- **Hardware**: Apple Silicon (M1/M2/M3) recommended
+- **Memory**: Minimum 4GB RAM
+- **Storage**: ~500MB for model files
+
+### Model Files
+- `AudioEncoder.mlmodelc/` - Audio encoder model
+- `MelSpectrogram.mlmodelc/` - Mel spectrogram processor
+- `TextDecoder.mlmodelc/` - Text decoder model
+- `*.mlcomputeplan.json` - Compute plans for optimization
+
+## Usage Examples
+
+### Basic Usage
+```python
+import whisperkit
+
+# Load model
+model = whisperkit.load_model("your-username/Breeze-ASR-25_coreml")
+
+# Transcribe audio file
+result = model.transcribe("audio.wav")
+print(result.text)
+```
+
+### Advanced Usage
+```python
+# With custom parameters
+result = model.transcribe(
+    "audio.wav",
+    language="en",
+    task="transcribe",
+    temperature=0.0
+)
+```
+
+## Citation
+
+```bibtex
+@article{breeze-asr-25-coreml,
+  title={Breeze-ASR-25 CoreML: On-Device Speech Recognition for Apple Silicon},
+  author={MediaTek Research},
+  journal={Hugging Face Model Hub},
+  year={2024}
+}
+```
+
+## Contact
+
+For questions or issues related to this model, please contact MediaTek Research or create an issue in the model repository.