Upload 4 files

TokenDurationPrediction.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d844856c854d42e6a58215dae5f75f82ea4da7cb7dbefb60db082a56c3a223dc
+size 243

TokenDurationPrediction.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03bd0964aae75139a64e2d25090b2c25c4aabe234bc5f63ae23d5e4d616d25d3
+size 424

TokenDurationPrediction.mlmodelc/metadata.json

@@ -0,0 +1,85 @@
+[
+  {
+    "shortDescription" : "Token and duration prediction for TDT decoder",
+    "metadataOutputVersion" : "3.0",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "var_17",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "reduce_max_0",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "var_24",
+        "type" : "MultiArray"
+      }
+    ],
+    "version" : "1.0",
+    "modelParameters" : [
+
+    ],
+    "author" : "FluidAudio",
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "SliceByIndex" : 2,
+      "Ios16.reduceArgmax" : 2,
+      "Ios16.reshape" : 1,
+      "Ios16.reduceMax" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "stateSchema" : [
+
+    ],
+    "isUpdatable" : "0",
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1030)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1, 1030]",
+        "name" : "logits",
+        "type" : "MultiArray"
+      }
+    ],
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "com.github.apple.coremltools.version" : "8.3.0"
+    },
+    "generatedClassName" : "TokenDurationPrediction",
+    "method" : "predict"
+  }
+]

TokenDurationPrediction.mlmodelc/model.mil

@@ -0,0 +1,25 @@
+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, [1, 1, 1, 1030]> logits) {
+            tensor<int32, [1]> var_3 = const()[name = tensor<string, []>("op_3"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1030]> flattened_cast_fp16 = reshape(shape = var_3, x = logits)[name = tensor<string, []>("flattened_cast_fp16")];
+            tensor<int32, [1]> token_logits_begin_0 = const()[name = tensor<string, []>("token_logits_begin_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1]> token_logits_end_0 = const()[name = tensor<string, []>("token_logits_end_0"), val = tensor<int32, [1]>([1025])];
+            tensor<bool, [1]> token_logits_end_mask_0 = const()[name = tensor<string, []>("token_logits_end_mask_0"), val = tensor<bool, [1]>([false])];
+            tensor<fp16, [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 = flattened_cast_fp16)[name = tensor<string, []>("token_logits_cast_fp16")];
+            tensor<int32, [1]> duration_logits_begin_0 = const()[name = tensor<string, []>("duration_logits_begin_0"), val = tensor<int32, [1]>([1025])];
+            tensor<int32, [1]> duration_logits_end_0 = const()[name = tensor<string, []>("duration_logits_end_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1]> duration_logits_end_mask_0 = const()[name = tensor<string, []>("duration_logits_end_mask_0"), val = tensor<bool, [1]>([true])];
+            tensor<fp16, [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 = flattened_cast_fp16)[name = tensor<string, []>("duration_logits_cast_fp16")];
+            tensor<int32, []> var_17_axis_0 = const()[name = tensor<string, []>("op_17_axis_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> var_17_keep_dims_0 = const()[name = tensor<string, []>("op_17_keep_dims_0"), val = tensor<bool, []>(true)];
+            tensor<int32, [1]> var_17 = reduce_argmax(axis = var_17_axis_0, keep_dims = var_17_keep_dims_0, x = token_logits_cast_fp16)[name = tensor<string, []>("op_17_cast_fp16")];
+            tensor<int32, [1]> reduce_max_0_axes_0 = const()[name = tensor<string, []>("reduce_max_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<bool, []> reduce_max_0_keep_dims_0 = const()[name = tensor<string, []>("reduce_max_0_keep_dims_0"), val = tensor<bool, []>(true)];
+            tensor<fp16, [1]> reduce_max_0 = reduce_max(axes = reduce_max_0_axes_0, keep_dims = reduce_max_0_keep_dims_0, x = token_logits_cast_fp16)[name = tensor<string, []>("reduce_max_0_cast_fp16")];
+            tensor<int32, []> var_24_axis_0 = const()[name = tensor<string, []>("op_24_axis_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> var_24_keep_dims_0 = const()[name = tensor<string, []>("op_24_keep_dims_0"), val = tensor<bool, []>(true)];
+            tensor<int32, [1]> var_24 = reduce_argmax(axis = var_24_axis_0, keep_dims = var_24_keep_dims_0, x = duration_logits_cast_fp16)[name = tensor<string, []>("op_24_cast_fp16")];
+        } -> (var_17, reduce_max_0, var_24);
+}