Add files using upload-large-folder tool

README.md

@@ -0,0 +1,26 @@
+---
+tags:
+- coreml
+- automatic-speech-recognition
+- parakeet
+- macos
+license: apache-2.0
+---
+
+# scriptrs models
+
+Runtime model bundle for `scriptrs`.
+
+Current contents:
+- Parakeet TDT v2 CoreML split models
+- vocabulary file
+- Silero VAD CoreML model for the `long-form` feature
+
+Layout:
+- `parakeet-v2/encoder.mlmodelc`
+- `parakeet-v2/decoder.mlmodelc`
+- `parakeet-v2/joint-decision.mlmodelc`
+- `parakeet-v2/vocab.txt`
+- `vad/silero-vad.mlmodelc`
+
+This repo is intended for `scriptrs` model download via the `online` feature.

parakeet-v2/config.json

@@ -0,0 +1,5 @@
+{
+    "model_type": "nemo-conformer-tdt",
+    "features_size": 128,
+    "subsampling_factor": 8
+}

parakeet-v2/decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46de1a6fe2e49d19a2125bc91acf020df7f2aea84ba821532aade8427a440b05
+size 243

parakeet-v2/decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d200ca07694a347f6d02a3886a062ae839831e094e443222f2e48a14945966a8
+size 554

parakeet-v2/decoder.mlmodelc/metadata.json

@@ -0,0 +1,123 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet decoder (RNNT prediction network)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 640 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 640, 1]",
+        "name" : "decoder",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 2 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[2, 1, 640]",
+        "name" : "h_out",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 2 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[2, 1, 640]",
+        "name" : "c_out",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Select" : 1,
+      "Ios17.squeeze" : 4,
+      "Ios17.gather" : 1,
+      "Ios17.cast" : 8,
+      "Ios17.lstm" : 2,
+      "Split" : 2,
+      "Ios17.add" : 1,
+      "Ios17.transpose" : 2,
+      "Ios17.greaterEqual" : 1,
+      "Identity" : 1,
+      "Stack" : 2
+    },
+    "computePrecision" : "Mixed (Float16, Float32, 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"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1]",
+        "name" : "targets",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "target_length",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 2 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[2, 1, 640]",
+        "name" : "h_in",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 2 × 1 × 640)",
+        "shortDescription" : "",
+        "shape" : "[2, 1, 640]",
+        "name" : "c_in",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-09-25",
+      "com.github.apple.coremltools.source" : "torch==2.7.0",
+      "com.github.apple.coremltools.version" : "9.0b1",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_decoder",
+    "method" : "predict"
+  }
+]

parakeet-v2/decoder.mlmodelc/model.mil

@@ -0,0 +1,73 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0b1"}})]
+{
+    func main<ios17>(tensor<fp32, [2, 1, 640]> c_in, tensor<fp32, [2, 1, 640]> h_in, tensor<int32, [1]> target_length, tensor<int32, [1, 1]> targets) {
+            tensor<int32, []> y_batch_dims_0 = const()[name = tensor<string, []>("y_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> y_validate_indices_0 = const()[name = tensor<string, []>("y_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1025, 640]> module_prediction_embed_weight_to_fp16 = const()[name = tensor<string, []>("module_prediction_embed_weight_to_fp16"), val = tensor<fp16, [1025, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<string, []> targets_to_int16_dtype_0 = const()[name = tensor<string, []>("targets_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> greater_equal_0_y_0 = const()[name = tensor<string, []>("greater_equal_0_y_0"), val = tensor<int32, []>(0)];
+            tensor<int16, [1, 1]> targets_to_int16 = cast(dtype = targets_to_int16_dtype_0, x = targets)[name = tensor<string, []>("cast_9")];
+            tensor<int32, [1, 1]> cast_1 = cast(dtype = cast_1_dtype_0, x = targets_to_int16)[name = tensor<string, []>("cast_8")];
+            tensor<bool, [1, 1]> greater_equal_0 = greater_equal(x = cast_1, y = greater_equal_0_y_0)[name = tensor<string, []>("greater_equal_0")];
+            tensor<int32, []> slice_by_index_0 = const()[name = tensor<string, []>("slice_by_index_0"), val = tensor<int32, []>(1025)];
+            tensor<int32, [1, 1]> add_2 = add(x = cast_1, y = slice_by_index_0)[name = tensor<string, []>("add_2")];
+            tensor<int32, [1, 1]> select_0 = select(a = cast_1, b = add_2, cond = greater_equal_0)[name = tensor<string, []>("select_0")];
+            tensor<int32, []> y_cast_fp16_cast_uint16_axis_0 = const()[name = tensor<string, []>("y_cast_fp16_cast_uint16_axis_0"), val = tensor<int32, []>(0)];
+            tensor<string, []> select_0_to_int16_dtype_0 = const()[name = tensor<string, []>("select_0_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1]> select_0_to_int16 = cast(dtype = select_0_to_int16_dtype_0, x = select_0)[name = tensor<string, []>("cast_7")];
+            tensor<fp16, [1, 1, 640]> y_cast_fp16_cast_uint16_cast_uint16 = gather(axis = y_cast_fp16_cast_uint16_axis_0, batch_dims = y_batch_dims_0, indices = select_0_to_int16, validate_indices = y_validate_indices_0, x = module_prediction_embed_weight_to_fp16)[name = tensor<string, []>("y_cast_fp16_cast_uint16_cast_uint16")];
+            tensor<int32, [3]> input_3_perm_0 = const()[name = tensor<string, []>("input_3_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<int32, []> split_0_num_splits_0 = const()[name = tensor<string, []>("split_0_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> split_0_axis_0 = const()[name = tensor<string, []>("split_0_axis_0"), val = tensor<int32, []>(0)];
+            tensor<string, []> h_in_to_fp16_dtype_0 = const()[name = tensor<string, []>("h_in_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [2, 1, 640]> h_in_to_fp16 = cast(dtype = h_in_to_fp16_dtype_0, x = h_in)[name = tensor<string, []>("cast_6")];
+            tensor<fp16, [1, 1, 640]> split_0_cast_fp16_0, tensor<fp16, [1, 1, 640]> split_0_cast_fp16_1 = split(axis = split_0_axis_0, num_splits = split_0_num_splits_0, x = h_in_to_fp16)[name = tensor<string, []>("split_0_cast_fp16")];
+            tensor<int32, []> split_1_num_splits_0 = const()[name = tensor<string, []>("split_1_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> split_1_axis_0 = const()[name = tensor<string, []>("split_1_axis_0"), val = tensor<int32, []>(0)];
+            tensor<string, []> c_in_to_fp16_dtype_0 = const()[name = tensor<string, []>("c_in_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [2, 1, 640]> c_in_to_fp16 = cast(dtype = c_in_to_fp16_dtype_0, x = c_in)[name = tensor<string, []>("cast_5")];
+            tensor<fp16, [1, 1, 640]> split_1_cast_fp16_0, tensor<fp16, [1, 1, 640]> split_1_cast_fp16_1 = split(axis = split_1_axis_0, num_splits = split_1_num_splits_0, x = c_in_to_fp16)[name = tensor<string, []>("split_1_cast_fp16")];
+            tensor<int32, [1]> input_lstm_layer_0_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("input_lstm_layer_0_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_lstm_layer_0_lstm_h0_squeeze_cast_fp16 = squeeze(axes = input_lstm_layer_0_lstm_h0_squeeze_axes_0, x = split_0_cast_fp16_0)[name = tensor<string, []>("input_lstm_layer_0_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> input_lstm_layer_0_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("input_lstm_layer_0_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_lstm_layer_0_lstm_c0_squeeze_cast_fp16 = squeeze(axes = input_lstm_layer_0_lstm_c0_squeeze_axes_0, x = split_1_cast_fp16_0)[name = tensor<string, []>("input_lstm_layer_0_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> input_lstm_layer_0_direction_0 = const()[name = tensor<string, []>("input_lstm_layer_0_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> input_lstm_layer_0_output_sequence_0 = const()[name = tensor<string, []>("input_lstm_layer_0_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_lstm_layer_0_recurrent_activation_0 = const()[name = tensor<string, []>("input_lstm_layer_0_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_lstm_layer_0_cell_activation_0 = const()[name = tensor<string, []>("input_lstm_layer_0_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_lstm_layer_0_activation_0 = const()[name = tensor<string, []>("input_lstm_layer_0_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [2560, 640]> concat_1_to_fp16 = const()[name = tensor<string, []>("concat_1_to_fp16"), val = tensor<fp16, [2560, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1312128)))];
+            tensor<fp16, [2560, 640]> concat_2_to_fp16 = const()[name = tensor<string, []>("concat_2_to_fp16"), val = tensor<fp16, [2560, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4588992)))];
+            tensor<fp16, [2560]> concat_0_to_fp16 = const()[name = tensor<string, []>("concat_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7865856)))];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = y_cast_fp16_cast_uint16_cast_uint16)[name = tensor<string, []>("transpose_2")];
+            tensor<fp16, [1, 1, 640]> input_lstm_layer_0_cast_fp16_0, tensor<fp16, [1, 640]> input_lstm_layer_0_cast_fp16_1, tensor<fp16, [1, 640]> input_lstm_layer_0_cast_fp16_2 = lstm(activation = input_lstm_layer_0_activation_0, bias = concat_0_to_fp16, cell_activation = input_lstm_layer_0_cell_activation_0, direction = input_lstm_layer_0_direction_0, initial_c = input_lstm_layer_0_lstm_c0_squeeze_cast_fp16, initial_h = input_lstm_layer_0_lstm_h0_squeeze_cast_fp16, output_sequence = input_lstm_layer_0_output_sequence_0, recurrent_activation = input_lstm_layer_0_recurrent_activation_0, weight_hh = concat_2_to_fp16, weight_ih = concat_1_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("input_lstm_layer_0_cast_fp16")];
+            tensor<int32, [1]> input_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("input_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_lstm_h0_squeeze_cast_fp16 = squeeze(axes = input_lstm_h0_squeeze_axes_0, x = split_0_cast_fp16_1)[name = tensor<string, []>("input_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> input_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("input_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_lstm_c0_squeeze_cast_fp16 = squeeze(axes = input_lstm_c0_squeeze_axes_0, x = split_1_cast_fp16_1)[name = tensor<string, []>("input_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> input_direction_0 = const()[name = tensor<string, []>("input_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> input_output_sequence_0 = const()[name = tensor<string, []>("input_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_recurrent_activation_0 = const()[name = tensor<string, []>("input_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_cell_activation_0 = const()[name = tensor<string, []>("input_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_activation_0 = const()[name = tensor<string, []>("input_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [2560, 640]> concat_4_to_fp16 = const()[name = tensor<string, []>("concat_4_to_fp16"), val = tensor<fp16, [2560, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7871040)))];
+            tensor<fp16, [2560, 640]> concat_5_to_fp16 = const()[name = tensor<string, []>("concat_5_to_fp16"), val = tensor<fp16, [2560, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11147904)))];
+            tensor<fp16, [2560]> concat_3_to_fp16 = const()[name = tensor<string, []>("concat_3_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14424768)))];
+            tensor<fp16, [1, 1, 640]> input_cast_fp16_0, tensor<fp16, [1, 640]> input_cast_fp16_1, tensor<fp16, [1, 640]> input_cast_fp16_2 = lstm(activation = input_activation_0, bias = concat_3_to_fp16, cell_activation = input_cell_activation_0, direction = input_direction_0, initial_c = input_lstm_c0_squeeze_cast_fp16, initial_h = input_lstm_h0_squeeze_cast_fp16, output_sequence = input_output_sequence_0, recurrent_activation = input_recurrent_activation_0, weight_hh = concat_5_to_fp16, weight_ih = concat_4_to_fp16, x = input_lstm_layer_0_cast_fp16_0)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<int32, []> obj_3_axis_0 = const()[name = tensor<string, []>("obj_3_axis_0"), val = tensor<int32, []>(0)];
+            tensor<fp16, [2, 1, 640]> obj_3_cast_fp16 = stack(axis = obj_3_axis_0, values = (input_lstm_layer_0_cast_fp16_1, input_cast_fp16_1))[name = tensor<string, []>("obj_3_cast_fp16")];
+            tensor<string, []> obj_3_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("obj_3_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, []> obj_axis_0 = const()[name = tensor<string, []>("obj_axis_0"), val = tensor<int32, []>(0)];
+            tensor<fp16, [2, 1, 640]> obj_cast_fp16 = stack(axis = obj_axis_0, values = (input_lstm_layer_0_cast_fp16_2, input_cast_fp16_2))[name = tensor<string, []>("obj_cast_fp16")];
+            tensor<string, []> obj_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("obj_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, [3]> transpose_0_perm_0 = const()[name = tensor<string, []>("transpose_0_perm_0"), val = tensor<int32, [3]>([1, 2, 0])];
+            tensor<string, []> transpose_0_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("transpose_0_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp16, [1, 640, 1]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = input_cast_fp16_0)[name = tensor<string, []>("transpose_1")];
+            tensor<fp32, [1, 640, 1]> decoder = cast(dtype = transpose_0_cast_fp16_to_fp32_dtype_0, x = transpose_0_cast_fp16)[name = tensor<string, []>("cast_2")];
+            tensor<fp32, [2, 1, 640]> c_out = cast(dtype = obj_cast_fp16_to_fp32_dtype_0, x = obj_cast_fp16)[name = tensor<string, []>("cast_3")];
+            tensor<fp32, [2, 1, 640]> h_out = cast(dtype = obj_3_cast_fp16_to_fp32_dtype_0, x = obj_3_cast_fp16)[name = tensor<string, []>("cast_4")];
+            tensor<int32, [1]> target_length_tmp = identity(x = target_length)[name = tensor<string, []>("target_length_tmp")];
+        } -> (decoder, h_out, c_out);
+}

parakeet-v2/decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27d26890221d82322c1092fd99d7b40578e435d5cf4b83c887c42603caf97aba
+size 14429952

parakeet-v2/encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42e638870d73f26b332918a3496ce36793fbb413a81cbd3d16ba01328637a105
+size 243

parakeet-v2/encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4def7aa848599ad0e17a8b9a982edcdbf33cf92e1f4b798de32e2ca0bc74b030
+size 485

parakeet-v2/encoder.mlmodelc/metadata.json

@@ -0,0 +1,106 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "enc6bit-palettize quantized - encoder",
+    "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, Palettized (6 bits))",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.floor" : 3,
+      "Ios17.logicalAnd" : 2,
+      "Ios17.reshape" : 145,
+      "Ios16.softmax" : 24,
+      "Ios17.matmul" : 72,
+      "Ios17.transpose" : 172,
+      "Split" : 24,
+      "Ios17.expandDims" : 5,
+      "Select" : 72,
+      "Ios17.add" : 174,
+      "Tile" : 1,
+      "Ios17.sliceByIndex" : 48,
+      "Ios16.sigmoid" : 24,
+      "Pad" : 48,
+      "Ios17.logicalNot" : 2,
+      "Ios17.layerNorm" : 120,
+      "Ios16.silu" : 72,
+      "Ios17.less" : 1,
+      "Ios16.constexprLutToDense" : 294,
+      "Ios17.conv" : 77,
+      "Ios16.relu" : 3,
+      "Ios17.linear" : 193,
+      "Ios17.cast" : 4,
+      "Ios17.mul" : 99
+    },
+    "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"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 128 × 1501)",
+        "shortDescription" : "",
+        "shape" : "[1, 128, 1501]",
+        "name" : "mel",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "mel_length",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-09-25",
+      "com.github.apple.coremltools.source" : "torch==2.7.0",
+      "com.github.apple.coremltools.version" : "9.0b1",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_encoder",
+    "method" : "predict"
+  }
+]

parakeet-v2/encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4adc7ad44f9d05e1bffeb2b06d3bb02861a5c7602dff63a6b494aed3bf8a6c3e
+size 445187200

parakeet-v2/joint-decision.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1183ba213bb94a918c8d2cad19ab045320618f97f6ca662245b3936d7b090f7
+size 243

parakeet-v2/joint-decision.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2c6752f1c8cf2d3f6f26ec93195c9bfa759ad59edf9f806696a138154f96f11
+size 534

parakeet-v2/joint-decision.mlmodelc/metadata.json

@@ -0,0 +1,103 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "shortDescription" : "Parakeet single-step joint decision (current frame)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1]",
+        "name" : "token_id",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1]",
+        "name" : "token_prob",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1]",
+        "name" : "duration",
+        "type" : "MultiArray"
+      }
+    ],
+    "storagePrecision" : "Float16",
+    "modelParameters" : [
+
+    ],
+    "author" : "Fluid Inference",
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.reduceArgmax" : 2,
+      "Ios17.squeeze" : 1,
+      "Ios17.cast" : 4,
+      "Ios17.linear" : 3,
+      "Ios17.transpose" : 2,
+      "Ios17.sliceByIndex" : 2,
+      "Ios17.add" : 1,
+      "Ios16.relu" : 1,
+      "Ios16.softmax" : 1,
+      "Ios17.gatherAlongAxis" : 1,
+      "Ios17.expandDims" : 3
+    },
+    "computePrecision" : "Mixed (Float16, Float32, 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"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 1024 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1024, 1]",
+        "name" : "encoder_step",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 640 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 640, 1]",
+        "name" : "decoder_step",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-09-25",
+      "com.github.apple.coremltools.source" : "torch==2.7.0",
+      "com.github.apple.coremltools.version" : "9.0b1",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "generatedClassName" : "parakeet_joint_decision_single_step",
+    "method" : "predict"
+  }
+]

parakeet-v2/joint-decision.mlmodelc/model.mil

@@ -0,0 +1,58 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0b1"}})]
+{
+    func main<ios17>(tensor<fp32, [1, 640, 1]> decoder_step, tensor<fp32, [1, 1024, 1]> encoder_step) {
+            tensor<int32, [3]> input_1_perm_0 = const()[name = tensor<string, []>("input_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> encoder_step_to_fp16_dtype_0 = const()[name = tensor<string, []>("encoder_step_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<int32, [3]> input_3_perm_0 = const()[name = tensor<string, []>("input_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> decoder_step_to_fp16_dtype_0 = const()[name = tensor<string, []>("decoder_step_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [640, 1024]> joint_module_enc_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_enc_weight_to_fp16"), val = tensor<fp16, [640, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [640]> joint_module_enc_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_enc_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1310848)))];
+            tensor<fp16, [1, 1024, 1]> encoder_step_to_fp16 = cast(dtype = encoder_step_to_fp16_dtype_0, x = encoder_step)[name = tensor<string, []>("cast_3")];
+            tensor<fp16, [1, 1, 1024]> input_1_cast_fp16 = transpose(perm = input_1_perm_0, x = encoder_step_to_fp16)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 1, 640]> linear_0_cast_fp16 = linear(bias = joint_module_enc_bias_to_fp16, weight = joint_module_enc_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<fp16, [640, 640]> joint_module_pred_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_pred_weight_to_fp16"), val = tensor<fp16, [640, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1312192)))];
+            tensor<fp16, [640]> joint_module_pred_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2131456)))];
+            tensor<fp16, [1, 640, 1]> decoder_step_to_fp16 = cast(dtype = decoder_step_to_fp16_dtype_0, x = decoder_step)[name = tensor<string, []>("cast_2")];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = decoder_step_to_fp16)[name = tensor<string, []>("transpose_0")];
+            tensor<fp16, [1, 1, 640]> linear_1_cast_fp16 = linear(bias = joint_module_pred_bias_to_fp16, weight = joint_module_pred_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
+            tensor<int32, [1]> var_23_axes_0 = const()[name = tensor<string, []>("op_23_axes_0"), val = tensor<int32, [1]>([2])];
+            tensor<fp16, [1, 1, 1, 640]> var_23_cast_fp16 = expand_dims(axes = var_23_axes_0, x = linear_0_cast_fp16)[name = tensor<string, []>("op_23_cast_fp16")];
+            tensor<int32, [1]> var_24_axes_0 = const()[name = tensor<string, []>("op_24_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, 1, 640]> var_24_cast_fp16 = expand_dims(axes = var_24_axes_0, x = linear_1_cast_fp16)[name = tensor<string, []>("op_24_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_5_cast_fp16 = add(x = var_23_cast_fp16, y = var_24_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_7_cast_fp16 = relu(x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
+            tensor<fp16, [1030, 640]> joint_module_joint_net_2_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_joint_net_2_weight_to_fp16"), val = tensor<fp16, [1030, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2132800)))];
+            tensor<fp16, [1030]> joint_module_joint_net_2_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_joint_net_2_bias_to_fp16"), val = tensor<fp16, [1030]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3451264)))];
+            tensor<fp16, [1, 1, 1, 1030]> linear_2_cast_fp16 = linear(bias = joint_module_joint_net_2_bias_to_fp16, weight = joint_module_joint_net_2_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<int32, [4]> token_logits_begin_0 = const()[name = tensor<string, []>("token_logits_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> token_logits_end_0 = const()[name = tensor<string, []>("token_logits_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1025])];
+            tensor<bool, [4]> token_logits_end_mask_0 = const()[name = tensor<string, []>("token_logits_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 1, 1, 1025]> token_logits_cast_fp16 = slice_by_index(begin = token_logits_begin_0, end = token_logits_end_0, end_mask = token_logits_end_mask_0, x = linear_2_cast_fp16)[name = tensor<string, []>("token_logits_cast_fp16")];
+            tensor<int32, [4]> duration_logits_begin_0 = const()[name = tensor<string, []>("duration_logits_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1025])];
+            tensor<int32, [4]> duration_logits_end_0 = const()[name = tensor<string, []>("duration_logits_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1030])];
+            tensor<bool, [4]> duration_logits_end_mask_0 = const()[name = tensor<string, []>("duration_logits_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 1, 1, 5]> duration_logits_cast_fp16 = slice_by_index(begin = duration_logits_begin_0, end = duration_logits_end_0, end_mask = duration_logits_end_mask_0, x = linear_2_cast_fp16)[name = tensor<string, []>("duration_logits_cast_fp16")];
+            tensor<int32, []> var_43_axis_0 = const()[name = tensor<string, []>("op_43_axis_0"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> var_43_keep_dims_0 = const()[name = tensor<string, []>("op_43_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_43_output_dtype_0 = const()[name = tensor<string, []>("op_43_output_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 1, 1]> token_id = reduce_argmax(axis = var_43_axis_0, keep_dims = var_43_keep_dims_0, output_dtype = var_43_output_dtype_0, x = token_logits_cast_fp16)[name = tensor<string, []>("op_43_cast_fp16")];
+            tensor<int32, []> var_49 = const()[name = tensor<string, []>("op_49"), val = tensor<int32, []>(-1)];
+            tensor<fp16, [1, 1, 1, 1025]> token_probs_all_cast_fp16 = softmax(axis = var_49, x = token_logits_cast_fp16)[name = tensor<string, []>("token_probs_all_cast_fp16")];
+            tensor<int32, [1]> var_58_axes_0 = const()[name = tensor<string, []>("op_58_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<int32, [1, 1, 1, 1]> var_58 = expand_dims(axes = var_58_axes_0, x = token_id)[name = tensor<string, []>("op_58")];
+            tensor<int32, []> var_59 = const()[name = tensor<string, []>("op_59"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> var_61_validate_indices_0 = const()[name = tensor<string, []>("op_61_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_58_to_int16_dtype_0 = const()[name = tensor<string, []>("op_58_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1, 1, 1]> var_58_to_int16 = cast(dtype = var_58_to_int16_dtype_0, x = var_58)[name = tensor<string, []>("cast_1")];
+            tensor<fp16, [1, 1, 1, 1]> var_61_cast_fp16_cast_int16 = gather_along_axis(axis = var_59, indices = var_58_to_int16, validate_indices = var_61_validate_indices_0, x = token_probs_all_cast_fp16)[name = tensor<string, []>("op_61_cast_fp16_cast_int16")];
+            tensor<int32, [1]> var_63_axes_0 = const()[name = tensor<string, []>("op_63_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, 1, 1]> var_63_cast_fp16 = squeeze(axes = var_63_axes_0, x = var_61_cast_fp16_cast_int16)[name = tensor<string, []>("op_63_cast_fp16")];
+            tensor<string, []> var_63_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_63_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, []> var_66_axis_0 = const()[name = tensor<string, []>("op_66_axis_0"), val = tensor<int32, []>(-1)];
+            tensor<bool, []> var_66_keep_dims_0 = const()[name = tensor<string, []>("op_66_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_66_output_dtype_0 = const()[name = tensor<string, []>("op_66_output_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 1, 1]> duration = reduce_argmax(axis = var_66_axis_0, keep_dims = var_66_keep_dims_0, output_dtype = var_66_output_dtype_0, x = duration_logits_cast_fp16)[name = tensor<string, []>("op_66_cast_fp16")];
+            tensor<fp32, [1, 1, 1]> token_prob = cast(dtype = var_63_cast_fp16_to_fp32_dtype_0, x = var_63_cast_fp16)[name = tensor<string, []>("cast_0")];
+        } -> (token_id, token_prob, duration);
+}

parakeet-v2/joint-decision.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca22a65903a05e64137677da608077578a8606090a598abf4875fa6199aaa19d
+size 3453388

parakeet-v2/vocab.txt

@@ -0,0 +1,1025 @@
+<unk> 0
+▁t 1
+▁th 2
+▁a 3
+in 4
+▁the 5
+re 6
+▁w 7
+▁o 8
+▁s 9
+at 10
+ou 11
+er 12
+nd 13
+▁i 14
+▁b 15
+▁c 16
+on 17
+▁h 18
+ing 19
+▁to 20
+▁m 21
+en 22
+▁f 23
+▁p 24
+an 25
+▁d 26
+es 27
+or 28
+ll 29
+▁of 30
+▁and 31
+▁y 32
+▁l 33
+▁I 34
+it 35
+▁in 36
+is 37
+ed 38
+▁g 39
+▁you 40
+ar 41
+▁that 42
+om 43
+as 44
+▁n 45
+ve 46
+us 47
+ic 48
+ow 49
+al 50
+▁it 51
+▁be 52
+▁wh 53
+le 54
+ion 55
+ut 56
+ot 57
+▁we 58
+▁is 59
+▁e 60
+et 61
+ay 62
+▁re 63
+▁on 64
+▁T 65
+▁A 66
+▁ha 67
+ent 68
+ke 69
+ct 70
+▁S 71
+ig 72
+ver 73
+▁Th 74
+all 75
+id 76
+▁for 77
+ro 78
+▁he 79
+se 80
+▁this 81
+ld 82
+ly 83
+▁go 84
+▁k 85
+▁st 86
+st 87
+ch 88
+▁li 89
+▁u 90
+am 91
+ur 92
+ce 93
+ith 94
+im 95
+▁so 96
+▁have 97
+▁do 98
+ht 99
+th 100
+▁an 101
+▁with 102
+ad 103
+▁r 104
+ir 105
+▁was 106
+▁as 107
+▁W 108
+▁are 109
+ust 110
+ally 111
+▁j 112
+▁se 113
+ation 114
+od 115
+ere 116
+▁like 117
+▁not 118
+▁kn 119
+ight 120
+▁B 121
+▁they 122
+▁And 123
+▁know 124
+ome 125
+op 126
+▁can 127
+▁or 128
+▁sh 129
+▁me 130
+ill 131
+ant 132
+ck 133
+▁what 134
+▁at 135
+▁ab 136
+ould 137
+ol 138
+▁So 139
+▁C 140
+use 141
+ter 142
+il 143
+▁but 144
+▁just 145
+▁ne 146
+▁de 147
+ra 148
+ore 149
+▁there 150
+ul 151
+out 152
+▁con 153
+▁all 154
+▁The 155
+ers 156
+▁H 157
+▁fr 158
+▁pro 159
+ge 160
+ea 161
+▁Y 162
+▁O 163
+▁M 164
+pp 165
+▁com 166
+ess 167
+▁ch 168
+▁al 169
+est 170
+ate 171
+qu 172
+▁lo 173
+▁ex 174
+very 175
+▁su 176
+ain 177
+▁one 178
+ca 179
+art 180
+ist 181
+if 182
+ive 183
+▁if 184
+ink 185
+nt 186
+ab 187
+▁about 188
+▁going 189
+▁v 190
+▁wor 191
+um 192
+ok 193
+▁your 194
+▁my 195
+ind 196
+▁get 197
+cause 198
+▁from 199
+▁don 200
+ri 201
+pe 202
+un 203
+ity 204
+▁up 205
+▁P 206
+▁out 207
+ort 208
+▁L 209
+ment 210
+el 211
+▁N 212
+▁some 213
+ich 214
+and 215
+▁think 216
+em 217
+oug 218
+▁G 219
+os 220
+▁D 221
+res 222
+▁because 223
+▁by 224
+ake 225
+▁int 226
+ie 227
+▁us 228
+▁tr 229
+▁then 230
+ack 231
+▁pl 232
+▁here 233
+▁pe 234
+her 235
+▁will 236
+▁F 237
+▁which 238
+ard 239
+▁right 240
+▁thing 241
+▁want 242
+ies 243
+ople 244
+▁It 245
+▁them 246
+ame 247
+▁We 248
+our 249
+▁say 250
+▁R 251
+▁people 252
+▁see 253
+▁who 254
+ast 255
+ure 256
+ect 257
+ear 258
+▁tim 259
+▁E 260
+▁You 261
+▁would 262
+▁when 263
+ven 264
+▁our 265
+ci 266
+▁really 267
+▁more 268
+ound 269
+ose 270
+ak 271
+▁co 272
+ide 273
+ough 274
+▁had 275
+so 276
+▁qu 277
+eah 278
+▁were 279
+ine 280
+▁act 281
+ther 282
+▁these 283
+▁how 284
+▁now 285
+▁sa 286
+ud 287
+▁Wh 288
+▁man 289
+ous 290
+one 291
+pt 292
+ff 293
+ong 294
+▁has 295
+▁any 296
+▁very 297
+▁But 298
+▁look 299
+iv 300
+itt 301
+▁time 302
+▁mo 303
+▁ar 304
+hing 305
+▁le 306
+▁work 307
+▁their 308
+are 309
+▁his 310
+per 311
+ions 312
+▁im 313
+▁ag 314
+▁J 315
+▁no 316
+▁en 317
+▁got 318
+ag 319
+▁sp 320
+ans 321
+act 322
+▁te 323
+▁also 324
+iz 325
+ice 326
+▁That 327
+▁cl 328
+▁been 329
+▁way 330
+▁fe 331
+▁did 332
+ple 333
+ually 334
+▁other 335
+▁U 336
+ite 337
+age 338
+omet 339
+ber 340
+reat 341
+ree 342
+▁into 343
+own 344
+▁tw 345
+▁part 346
+alk 347
+▁where 348
+▁need 349
+▁every 350
+pl 351
+▁ad 352
+ry 353
+▁over 354
+ble 355
+ap 356
+ue 357
+▁kind 358
+▁po 359
+▁back 360
+▁cont 361
+iff 362
+▁somet 363
+▁pr 364
+nder 365
+ire 366
+▁good 367
+▁than 368
+ace 369
+▁gu 370
+ep 371
+og 372
+ick 373
+way 374
+▁lot 375
+▁un 376
+▁things 377
+▁In 378
+ish 379
+kay 380
+▁well 381
+▁could 382
+▁pre 383
+▁two 384
+irst 385
+▁diff 386
+ach 387
+cc 388
+ittle 389
+int 390
+▁He 391
+▁those 392
+ence 393
+ip 394
+ase 395
+▁him 396
+▁make 397
+▁little 398
+ical 399
+▁gr 400
+▁year 401
+ass 402
+▁thr 403
+uch 404
+ated 405
+▁This 406
+▁off 407
+▁res 408
+ac 409
+ance 410
+▁actually 411
+▁talk 412
+ult 413
+able 414
+orm 415
+▁dis 416
+▁first 417
+ations 418
+▁something 419
+▁she 420
+sel 421
+▁let 422
+ord 423
+▁may 424
+ia 425
+▁am 426
+▁her 427
+▁said 428
+▁bo 429
+be 430
+ount 431
+▁much 432
+▁per 433
+▁even 434
+▁differe 435
+vel 436
+ary 437
+▁app 438
+ving 439
+▁comm 440
+▁imp 441
+ys 442
+▁again 443
+ress 444
+▁yeah 445
+▁down 446
+ang 447
+▁mean 448
+na 449
+ens 450
+▁does 451
+▁fo 452
+▁comp 453
+▁ro 454
+▁bl 455
+ody 456
+▁K 457
+▁through 458
+▁start 459
+uct 460
+▁only 461
+▁bet 462
+▁under 463
+▁br 464
+▁take 465
+ning 466
+▁bu 467
+▁use 468
+▁Ch 469
+xt 470
+co 471
+ory 472
+ild 473
+▁put 474
+▁call 475
+▁new 476
+other 477
+ting 478
+▁happ 479
+ater 480
+▁inc 481
+ition 482
+▁different 483
+▁should 484
+ade 485
+ign 486
+thing 487
+▁day 488
+fore 489
+▁Yeah 490
+ark 491
+ile 492
+ial 493
+▁come 494
+▁They 495
+��being 496
+▁try 497
+ious 498
+▁sc 499
+▁bit 500
+▁spe 501
+ub 502
+fe 503
+▁doing 504
+▁St 505
+vers 506
+av 507
+ty 508
+ian 509
+onna 510
+red 511
+wn 512
+▁ke 513
+form 514
+ors 515
+▁fl 516
+fter 517
+ail 518
+ents 519
+▁gonna 520
+▁point 521
+ces 522
+▁There 523
+self 524
+▁many 525
+▁If 526
+▁same 527
+▁sy 528
+▁quest 529
+▁most 530
+▁great 531
+▁What 532
+▁fu 533
+ug 534
+▁show 535
+we 536
+ual 537
+ons 538
+▁Be 539
+ically 540
+▁ser 541
+▁rem 542
+▁ind 543
+▁pers 544
+▁V 545
+he 546
+▁str 547
+ved 548
+▁still 549
+ank 550
+▁rec 551
+▁wr 552
+ought 553
+day 554
+ath 555
+▁end 556
+▁bas 557
+ft 558
+erm 559
+body 560
+ph 561
+ject 562
+ict 563
+▁play 564
+▁Is 565
+ates 566
+▁ph 567
+oth 568
+▁acc 569
+get 570
+▁years 571
+▁em 572
+▁id 573
+▁Oh 574
+ves 575
+ever 576
+▁inter 577
+▁rel 578
+▁before 579
+▁feel 580
+igh 581
+▁three 582
+iss 583
+▁des 584
+ne 585
+▁why 586
+▁uh 587
+▁To 588
+▁cons 589
+▁hel 590
+▁after 591
+ower 592
+urn 593
+▁okay 594
+▁long 595
+▁bel 596
+▁around 597
+ful 598
+te 599
+ise 600
+▁ob 601
+▁supp 602
+ady 603
+ange 604
+aking 605
+▁pos 606
+atch 607
+▁tra 608
+gr 609
+▁might 610
+ert 611
+▁help 612
+ost 613
+▁too 614
+cial 615
+▁world 616
+▁give 617
+ike 618
+▁Okay 619
+ways 620
+▁min 621
+ward 622
+ily 623
+▁gen 624
+▁find 625
+▁dec 626
+ular 627
+ob 628
+▁tell 629
+▁Now 630
+▁sm 631
+▁cour 632
+▁real 633
+cess 634
+nds 635
+▁big 636
+▁num 637
+ction 638
+▁add 639
+▁set 640
+▁um 641
+ood 642
+ible 643
+▁own 644
+▁life 645
+ities 646
+▁its 647
+▁God 648
+pect 649
+▁didn 650
+stem 651
+les 652
+uc 653
+ib 654
+ating 655
+olog 656
+▁person 657
+▁inv 658
+ably 659
+▁sure 660
+▁reg 661
+lic 662
+▁stu 663
+▁cr 664
+▁ev 665
+ments 666
+▁another 667
+▁la 668
+▁last 669
+▁sub 670
+▁att 671
+▁op 672
+▁inst 673
+▁sl 674
+▁happen 675
+▁rep 676
+▁import 677
+ific 678
+ix 679
+▁made 680
+▁ear 681
+▁ac 682
+▁def 683
+ute 684
+▁next 685
+ative 686
+▁form 687
+▁guys 688
+▁system 689
+ew 690
+▁able 691
+ied 692
+▁always 693
+ren 694
+erest 695
+▁As 696
+▁mod 697
+▁done 698
+ings 699
+▁love 700
+ism 701
+▁ask 702
+old 703
+ered 704
+▁trans 705
+▁count 706
+ility 707
+▁high 708
+▁fin 709
+▁prob 710
+▁pol 711
+▁exam 712
+▁pres 713
+▁maybe 714
+ell 715
+▁stud 716
+▁prod 717
+▁car 718
+ock 719
+▁used 720
+oy 721
+stand 722
+▁No 723
+▁mon 724
+ks 725
+▁interest 726
+▁ent 727
+ited 728
+▁sort 729
+▁For 730
+▁today 731
+ics 732
+▁vide 733
+▁bec 734
+▁Well 735
+▁Al 736
+▁important 737
+▁such 738
+▁run 739
+▁keep 740
+▁fact 741
+ata 742
+ss 743
+▁never 744
+ween 745
+▁stuff 746
+ract 747
+▁question 748
+als 749
+▁sim 750
+vern 751
+ather 752
+▁course 753
+▁Of 754
+oc 755
+ness 756
+arch 757
+ize 758
+▁All 759
+ense 760
+blem 761
+▁probably 762
+hip 763
+▁number 764
+ention 765
+▁saying 766
+▁commun 767
+▁An 768
+akes 769
+▁belie 770
+▁between 771
+▁better 772
+cus 773
+▁place 774
+▁gener 775
+▁ca 776
+▁ins 777
+▁ass 778
+cond 779
+cept 780
+ull 781
+▁understand 782
+▁fun 783
+▁thought 784
+gan 785
+iew 786
+cy 787
+ution 788
+ope 789
+ason 790
+▁problem 791
+▁doesn 792
+ational 793
+▁read 794
+▁trying 795
+▁sch 796
+▁el 797
+ah 798
+atter 799
+▁exper 800
+▁four 801
+▁ele 802
+▁cou 803
+ont 804
+▁called 805
+▁partic 806
+▁open 807
+▁gl 808
+▁everything 809
+▁eff 810
+▁getting 811
+▁ty 812
+▁Am 813
+▁Because 814
+ave 815
+▁met 816
+▁Like 817
+oney 818
+▁ 819
+e 820
+t 821
+o 822
+a 823
+n 824
+i 825
+s 826
+h 827
+r 828
+l 829
+d 830
+u 831
+c 832
+y 833
+m 834
+g 835
+w 836
+f 837
+p 838
+, 839
+b 840
+. 841
+k 842
+v 843
+' 844
+I 845
+T 846
+A 847
+S 848
+j 849
+x 850
+W 851
+B 852
+C 853
+? 854
+0 855
+O 856
+- 857
+M 858
+H 859
+Y 860
+q 861
+1 862
+P 863
+z 864
+L 865
+D 866
+N 867
+G 868
+F 869
+R 870
+E 871
+2 872
+J 873
+U 874
+: 875
+5 876
+9 877
+3 878
+K 879
+4 880
+V 881
+8 882
+6 883
+7 884
+! 885
+% 886
+Q 887
+$ 888
+Z 889
+X 890
+é 891
+/ 892
+í 893
+á 894
+£ 895
+ó 896
+ā 897
+ü 898
+ñ 899
+ö 900
+è 901
+ç 902
+à 903
+¿ 904
+μ 905
+π 906
+ä 907
+ú 908
+θ 909
+ã 910
+φ 911
+ī 912
+σ 913
+ê 914
+ρ 915
+â 916
+ô 917
+^ 918
+€ 919
+É 920
+ū 921
+Δ 922
+λ 923
+α 924
+τ 925
+æ 926
+а 927
+о 928
+ν 929
+î 930
+γ 931
+ψ 932
+ē 933
+т 934
+ß 935
+ω 936
+ï 937
+ć 938
+č 939
+ε 940
+е 941
+и 942
+ò 943
+р 944
+β 945
+ø 946
+ł 947
+δ 948
+η 949
+п 950
+ë 951
+н 952
+с 953
+š 954
+Ü 955
+å 956
+ń 957
+ś 958
+я 959
+đ 960
+л 961
+м 962
+Ö 963
+û 964
+ș 965
+в 966
+Á 967
+Ø 968
+ù 969
+ο 970
+ч 971
+ь 972
+ž 973
+Φ 974
+у 975
+ę 976
+ι 977
+б 978
+г 979
+к 980
+ő 981
+Ś 982
+Ω 983
+κ 984
+υ 985
+ì 986
+Č 987
+έ 988
+х 989
+ы 990
+Å 991
+Ç 992
+ż 993
+ί 994
+ζ 995
+χ 996
+э 997
+Æ 998
+Í 999
+õ 1000
+ě 1001
+ħ 1002
+Ł 1003
+œ 1004
+Ž 1005
+ț 1006
+Γ 1007
+П 1008
+д 1009
+з 1010
+ф 1011
+¡ 1012
+À 1013
+Î 1014
+Ā 1015
+ė 1016
+Š 1017
+ź 1018
+Κ 1019
+Ψ 1020
+ά 1021
+ξ 1022
+ό 1023
+<blk> 1024

vad/silero-vad.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f856f53009d001936ae46da67ba5edb8a3d28d4ff50ed38646d9c025cc5fbb35
+size 243

vad/silero-vad.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e30e28650a03afaa1bd20041362933235b455245468f533c031b683c29730f19
+size 459

vad/silero-vad.mlmodelc/metadata.json

@@ -0,0 +1,113 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "probability",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "h_out",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "c_out",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.sqrt" : 1,
+      "Ios17.transpose" : 1,
+      "Ios17.conv" : 6,
+      "Ios16.sigmoid" : 1,
+      "Ios17.sliceByIndex" : 2,
+      "Ios17.add" : 2,
+      "Ios17.lstm" : 1,
+      "Ios16.relu" : 5,
+      "Ios17.pow" : 2,
+      "Ios17.expandDims" : 3,
+      "Ios17.squeeze" : 3,
+      "Ios17.reshape" : 1,
+      "Pad" : 1
+    },
+    "computePrecision" : "Mixed (Float16, 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" : "2026-02-28",
+      "com.github.apple.coremltools.source" : "torch==2.10.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 576)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 576]",
+        "name" : "audio",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "h",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "c",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "silero_vad",
+    "method" : "predict"
+  }
+]

vad/silero-vad.mlmodelc/model.mil

@@ -0,0 +1,104 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<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<ios17>(tensor<fp16, [1, 1, 576]> audio, tensor<fp16, [1, 1, 128]> c, tensor<fp16, [1, 1, 128]> h) {
+            tensor<int32, [6]> input_1_pad_0 = const()[name = tensor<string, []>("input_1_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 64])];
+            tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("reflect")];
+            tensor<fp16, []> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1, 640]> input_1_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = input_1_mode_0, pad = input_1_pad_0, x = audio)[name = tensor<string, []>("input_1_cast_fp16")];
+            tensor<string, []> x_1_pad_type_0 = const()[name = tensor<string, []>("x_1_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> x_1_strides_0 = const()[name = tensor<string, []>("x_1_strides_0"), val = tensor<int32, [1]>([128])];
+            tensor<int32, [2]> x_1_pad_0 = const()[name = tensor<string, []>("x_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_1_dilations_0 = const()[name = tensor<string, []>("x_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_1_groups_0 = const()[name = tensor<string, []>("x_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [258, 1, 256]> stft_conv_weight_to_fp16 = const()[name = tensor<string, []>("stft_conv_weight_to_fp16"), val = tensor<fp16, [258, 1, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1, 258, 4]> x_1_cast_fp16 = conv(dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = stft_conv_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("x_1_cast_fp16")];
+            tensor<int32, [3]> var_42_begin_0 = const()[name = tensor<string, []>("op_42_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<int32, [3]> var_42_end_0 = const()[name = tensor<string, []>("op_42_end_0"), val = tensor<int32, [3]>([1, 129, 4])];
+            tensor<bool, [3]> var_42_end_mask_0 = const()[name = tensor<string, []>("op_42_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
+            tensor<fp16, [1, 129, 4]> var_42_cast_fp16 = slice_by_index(begin = var_42_begin_0, end = var_42_end_0, end_mask = var_42_end_mask_0, x = x_1_cast_fp16)[name = tensor<string, []>("op_42_cast_fp16")];
+            tensor<int32, [3]> var_57_begin_0 = const()[name = tensor<string, []>("op_57_begin_0"), val = tensor<int32, [3]>([0, 129, 0])];
+            tensor<int32, [3]> var_57_end_0 = const()[name = tensor<string, []>("op_57_end_0"), val = tensor<int32, [3]>([1, 258, 4])];
+            tensor<bool, [3]> var_57_end_mask_0 = const()[name = tensor<string, []>("op_57_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
+            tensor<fp16, [1, 129, 4]> var_57_cast_fp16 = slice_by_index(begin = var_57_begin_0, end = var_57_end_0, end_mask = var_57_end_mask_0, x = x_1_cast_fp16)[name = tensor<string, []>("op_57_cast_fp16")];
+            tensor<fp16, []> var_63_promoted_to_fp16 = const()[name = tensor<string, []>("op_63_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
+            tensor<fp16, [1, 129, 4]> var_64_cast_fp16 = pow(x = var_42_cast_fp16, y = var_63_promoted_to_fp16)[name = tensor<string, []>("op_64_cast_fp16")];
+            tensor<fp16, []> var_65_promoted_to_fp16 = const()[name = tensor<string, []>("op_65_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
+            tensor<fp16, [1, 129, 4]> var_66_cast_fp16 = pow(x = var_57_cast_fp16, y = var_65_promoted_to_fp16)[name = tensor<string, []>("op_66_cast_fp16")];
+            tensor<fp16, [1, 129, 4]> var_68_cast_fp16 = add(x = var_64_cast_fp16, y = var_66_cast_fp16)[name = tensor<string, []>("op_68_cast_fp16")];
+            tensor<fp16, []> var_70_to_fp16 = const()[name = tensor<string, []>("op_70_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
+            tensor<fp16, [1, 129, 4]> var_71_cast_fp16 = add(x = var_68_cast_fp16, y = var_70_to_fp16)[name = tensor<string, []>("op_71_cast_fp16")];
+            tensor<fp16, [1, 129, 4]> input_3_cast_fp16 = sqrt(x = var_71_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<string, []> input_5_pad_type_0 = const()[name = tensor<string, []>("input_5_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_5_pad_0 = const()[name = tensor<string, []>("input_5_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_5_strides_0 = const()[name = tensor<string, []>("input_5_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_5_dilations_0 = const()[name = tensor<string, []>("input_5_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_5_groups_0 = const()[name = tensor<string, []>("input_5_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [128, 129, 3]> encoder_0_weight_to_fp16 = const()[name = tensor<string, []>("encoder_0_weight_to_fp16"), val = tensor<fp16, [128, 129, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132224)))];
+            tensor<fp16, [128]> encoder_0_bias_to_fp16 = const()[name = tensor<string, []>("encoder_0_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231360)))];
+            tensor<fp16, [1, 128, 4]> input_5_cast_fp16 = conv(bias = encoder_0_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = encoder_0_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<fp16, [1, 128, 4]> input_7_cast_fp16 = relu(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]>([2])];
+            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, [64, 128, 3]> encoder_1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_1_weight_to_fp16"), val = tensor<fp16, [64, 128, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231680)))];
+            tensor<fp16, [64]> encoder_1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_1_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280896)))];
+            tensor<fp16, [1, 64, 2]> input_9_cast_fp16 = conv(bias = encoder_1_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 = encoder_1_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
+            tensor<fp16, [1, 64, 2]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
+            tensor<string, []> input_13_pad_type_0 = const()[name = tensor<string, []>("input_13_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_13_pad_0 = const()[name = tensor<string, []>("input_13_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_13_strides_0 = const()[name = tensor<string, []>("input_13_strides_0"), val = tensor<int32, [1]>([2])];
+            tensor<int32, [1]> input_13_dilations_0 = const()[name = tensor<string, []>("input_13_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_13_groups_0 = const()[name = tensor<string, []>("input_13_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [64, 64, 3]> encoder_2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_2_weight_to_fp16"), val = tensor<fp16, [64, 64, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281088)))];
+            tensor<fp16, [64]> encoder_2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_2_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305728)))];
+            tensor<fp16, [1, 64, 1]> input_13_cast_fp16 = conv(bias = encoder_2_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = encoder_2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
+            tensor<fp16, [1, 64, 1]> input_15_cast_fp16 = relu(x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
+            tensor<string, []> input_17_pad_type_0 = const()[name = tensor<string, []>("input_17_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_17_pad_0 = const()[name = tensor<string, []>("input_17_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_17_strides_0 = const()[name = tensor<string, []>("input_17_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_17_dilations_0 = const()[name = tensor<string, []>("input_17_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_17_groups_0 = const()[name = tensor<string, []>("input_17_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [128, 64, 3]> encoder_3_weight_to_fp16 = const()[name = tensor<string, []>("encoder_3_weight_to_fp16"), val = tensor<fp16, [128, 64, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305920)))];
+            tensor<fp16, [128]> encoder_3_bias_to_fp16 = const()[name = tensor<string, []>("encoder_3_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355136)))];
+            tensor<fp16, [1, 128, 1]> input_17_cast_fp16 = conv(bias = encoder_3_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = encoder_3_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
+            tensor<fp16, [1, 128, 1]> x_cast_fp16 = relu(x = input_17_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
+            tensor<int32, [3]> transpose_0_perm_0 = const()[name = tensor<string, []>("transpose_0_perm_0"), val = tensor<int32, [3]>([2, 0, 1])];
+            tensor<int32, [1]> var_138_batch_first_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("op_138_batch_first_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 128]> var_138_batch_first_lstm_h0_squeeze_cast_fp16 = squeeze(axes = var_138_batch_first_lstm_h0_squeeze_axes_0, x = h)[name = tensor<string, []>("op_138_batch_first_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> var_138_batch_first_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("op_138_batch_first_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 128]> var_138_batch_first_lstm_c0_squeeze_cast_fp16 = squeeze(axes = var_138_batch_first_lstm_c0_squeeze_axes_0, x = c)[name = tensor<string, []>("op_138_batch_first_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> var_138_batch_first_direction_0 = const()[name = tensor<string, []>("op_138_batch_first_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> var_138_batch_first_output_sequence_0 = const()[name = tensor<string, []>("op_138_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> var_138_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("op_138_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> var_138_batch_first_cell_activation_0 = const()[name = tensor<string, []>("op_138_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> var_138_batch_first_activation_0 = const()[name = tensor<string, []>("op_138_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [512, 128]> concat_1_to_fp16 = const()[name = tensor<string, []>("concat_1_to_fp16"), val = tensor<fp16, [512, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355456)))];
+            tensor<fp16, [512, 128]> concat_2_to_fp16 = const()[name = tensor<string, []>("concat_2_to_fp16"), val = tensor<fp16, [512, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486592)))];
+            tensor<fp16, [512]> concat_0_to_fp16 = const()[name = tensor<string, []>("concat_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617728)))];
+            tensor<fp16, [1, 1, 128]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = x_cast_fp16)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 1, 128]> var_138_batch_first_cast_fp16_0, tensor<fp16, [1, 128]> var_138_batch_first_cast_fp16_1, tensor<fp16, [1, 128]> var_138_batch_first_cast_fp16_2 = lstm(activation = var_138_batch_first_activation_0, bias = concat_0_to_fp16, cell_activation = var_138_batch_first_cell_activation_0, direction = var_138_batch_first_direction_0, initial_c = var_138_batch_first_lstm_c0_squeeze_cast_fp16, initial_h = var_138_batch_first_lstm_h0_squeeze_cast_fp16, output_sequence = var_138_batch_first_output_sequence_0, recurrent_activation = var_138_batch_first_recurrent_activation_0, weight_hh = concat_2_to_fp16, weight_ih = concat_1_to_fp16, x = transpose_0_cast_fp16)[name = tensor<string, []>("op_138_batch_first_cast_fp16")];
+            tensor<int32, [1]> h_out_axes_0 = const()[name = tensor<string, []>("h_out_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 128]> h_out = expand_dims(axes = h_out_axes_0, x = var_138_batch_first_cast_fp16_1)[name = tensor<string, []>("h_out_cast_fp16")];
+            tensor<int32, [1]> var_140_axes_0 = const()[name = tensor<string, []>("op_140_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 128]> c_out = expand_dims(axes = var_140_axes_0, x = var_138_batch_first_cast_fp16_2)[name = tensor<string, []>("op_140_cast_fp16")];
+            tensor<int32, [1]> var_145_axes_0 = const()[name = tensor<string, []>("op_145_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 128]> var_145_cast_fp16 = squeeze(axes = var_145_axes_0, x = h_out)[name = tensor<string, []>("op_145_cast_fp16")];
+            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, 128, 1]> input_21_cast_fp16 = expand_dims(axes = input_21_axes_0, x = var_145_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
+            tensor<fp16, [1, 128, 1]> input_cast_fp16 = relu(x = input_21_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<string, []> var_159_pad_type_0 = const()[name = tensor<string, []>("op_159_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_159_strides_0 = const()[name = tensor<string, []>("op_159_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_159_pad_0 = const()[name = tensor<string, []>("op_159_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_159_dilations_0 = const()[name = tensor<string, []>("op_159_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_159_groups_0 = const()[name = tensor<string, []>("op_159_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 128, 1]> decoder_conv_weight_to_fp16 = const()[name = tensor<string, []>("decoder_conv_weight_to_fp16"), val = tensor<fp16, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618816)))];
+            tensor<fp16, [1]> decoder_conv_bias_to_fp16 = const()[name = tensor<string, []>("decoder_conv_bias_to_fp16"), val = tensor<fp16, [1]>([-0x1.868p-3])];
+            tensor<fp16, [1, 1, 1]> var_159_cast_fp16 = conv(bias = decoder_conv_bias_to_fp16, dilations = var_159_dilations_0, groups = var_159_groups_0, pad = var_159_pad_0, pad_type = var_159_pad_type_0, strides = var_159_strides_0, weight = decoder_conv_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("op_159_cast_fp16")];
+            tensor<fp16, [1, 1, 1]> d_cast_fp16 = sigmoid(x = var_159_cast_fp16)[name = tensor<string, []>("d_cast_fp16")];
+            tensor<int32, [1]> var_162 = const()[name = tensor<string, []>("op_162"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1]> probability = reshape(shape = var_162, x = d_cast_fp16)[name = tensor<string, []>("op_163_cast_fp16")];
+        } -> (probability, h_out, c_out);
+}

vad/silero-vad.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06ff3bff3db86d66b33836271da001e80e5ce38a28407b9c61e976ae9c713acf
+size 619136