Sortformer models for argmax-sdk-swift-alpha-2.0.1

sortformer/v2-1/384_94MB/AudioConformerPreEncoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:449a41e2580a2184a38661d236a2375838c45d64629d18639eeefe0ab22db783
+size 243

sortformer/v2-1/384_94MB/AudioConformerPreEncoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1734fffd4068cd4adc74eff589b0f57e48d930b848b9533e83c24bd02b572c2
+size 423

sortformer/v2-1/384_94MB/AudioConformerPreEncoder.mlmodelc/metadata.json

@@ -0,0 +1,64 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 385)",
+        "shortDescription" : "",
+        "shape" : "[1, 512, 1, 385]",
+        "name" : "downsampled_melspectrogram_features",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios18.transpose" : 1,
+      "Ios18.relu" : 3,
+      "Ios18.reshape" : 1,
+      "Ios18.conv" : 6
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-12",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "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 × 3073 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 3073, 128]",
+        "name" : "melspectrogram_features",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "AudioConformerPreEncoder",
+    "method" : "predict"
+  }
+]

sortformer/v2-1/384_94MB/AudioConformerPreEncoder.mlmodelc/model.mil

@@ -0,0 +1,61 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3404.16.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios18>(tensor<fp16, [1, 1, 3073, 128]> melspectrogram_features) {
+            string input_1_pad_type_0 = const()[name = string("input_1_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_1_pad_0 = const()[name = string("input_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_1_strides_0 = const()[name = string("input_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> input_1_dilations_0 = const()[name = string("input_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_1_groups_0 = const()[name = string("input_1_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 1, 3, 3]> conv_0_weight_to_fp16 = const()[name = string("conv_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
+            tensor<fp16, [256]> conv_0_bias_to_fp16 = const()[name = string("conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
+            tensor<fp16, [1, 256, 1537, 64]> input_1_cast_fp16 = conv(bias = conv_0_bias_to_fp16, dilations = input_1_dilations_0, groups = input_1_groups_0, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = input_1_strides_0, weight = conv_0_weight_to_fp16, x = melspectrogram_features)[name = string("input_1_cast_fp16")];
+            tensor<fp16, [1, 256, 1537, 64]> input_3_cast_fp16 = relu(x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
+            string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> conv_2_weight_to_fp16 = const()[name = string("conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
+            tensor<fp16, [256]> conv_2_bias_to_fp16 = const()[name = string("conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
+            tensor<fp16, [1, 256, 769, 32]> input_5_cast_fp16 = conv(bias = conv_2_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 = conv_2_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
+            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> conv_3_weight_to_fp16 = const()[name = string("conv_3_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
+            tensor<fp16, [256]> conv_3_bias_to_fp16 = const()[name = string("conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
+            tensor<fp16, [1, 256, 769, 32]> input_7_cast_fp16 = conv(bias = conv_3_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = conv_3_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
+            tensor<fp16, [1, 256, 769, 32]> input_9_cast_fp16 = relu(x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
+            string input_11_pad_type_0 = const()[name = string("input_11_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_11_strides_0 = const()[name = string("input_11_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_11_groups_0 = const()[name = string("input_11_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_11_dilations_0 = const()[name = string("input_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> conv_5_weight_to_fp16 = const()[name = string("conv_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
+            tensor<fp16, [256]> conv_5_bias_to_fp16 = const()[name = string("conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
+            tensor<fp16, [1, 256, 385, 16]> input_11_cast_fp16 = conv(bias = conv_5_bias_to_fp16, dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = conv_5_weight_to_fp16, x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
+            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> conv_6_weight_to_fp16 = const()[name = string("conv_6_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
+            tensor<fp16, [256]> conv_6_bias_to_fp16 = const()[name = string("conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
+            tensor<fp16, [1, 256, 385, 16]> input_13_cast_fp16 = conv(bias = conv_6_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 = conv_6_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
+            tensor<fp16, [1, 256, 385, 16]> x_cast_fp16 = relu(x = input_13_cast_fp16)[name = string("x_cast_fp16")];
+            tensor<int32, [4]> var_69_perm_0 = const()[name = string("op_69_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
+            tensor<int32, [4]> var_73 = const()[name = string("op_73"), val = tensor<int32, [4]>([1, 4096, 1, 385])];
+            tensor<fp16, [1, 256, 16, 385]> var_69_cast_fp16 = transpose(perm = var_69_perm_0, x = x_cast_fp16)[name = string("transpose_0")];
+            tensor<fp16, [1, 4096, 1, 385]> input_cast_fp16 = reshape(shape = var_73, x = var_69_cast_fp16)[name = string("input_cast_fp16")];
+            string var_85_pad_type_0 = const()[name = string("op_85_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> var_85_strides_0 = const()[name = string("op_85_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> var_85_pad_0 = const()[name = string("op_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> var_85_dilations_0 = const()[name = string("op_85_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 var_85_groups_0 = const()[name = string("op_85_groups_0"), val = int32(1)];
+            tensor<fp16, [512, 4096, 1, 1]> out_weight_to_fp16 = const()[name = string("out_weight_to_fp16"), val = tensor<fp16, [512, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
+            tensor<fp16, [512]> out_bias_to_fp16 = const()[name = string("out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4473600)))];
+            tensor<fp16, [1, 512, 1, 385]> downsampled_melspectrogram_features = conv(bias = out_bias_to_fp16, dilations = var_85_dilations_0, groups = var_85_groups_0, pad = var_85_pad_0, pad_type = var_85_pad_type_0, strides = var_85_strides_0, weight = out_weight_to_fp16, x = input_cast_fp16)[name = string("op_85_cast_fp16")];
+        } -> (downsampled_melspectrogram_features);
+}

sortformer/v2-1/384_94MB/AudioConformerPreEncoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:232fce1518edd08b1fcb0b39add3e7307c8c3ca3e6afa170f6937c57617d38e1
+size 4474688

sortformer/v2-1/384_94MB/LICENSE_NOTICE.txt

@@ -0,0 +1,7 @@
+Argmax proprietary and confidential. Under NDA.
+
+Copyright 2026 Argmax, Inc. All rights reserved.
+
+Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited.
+
+Please contact Argmax for licensing information at info@argmaxinc.com.

sortformer/v2-1/384_94MB/MelSpectrogram.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ed0dd5b53602d2a0037b2a4dff357301daeb133c6325f38616d197b2271afbc
+size 243

sortformer/v2-1/384_94MB/MelSpectrogram.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f4d7e4c41f2d9af18f6825078b5d71641606f5996e2bd553f4d47448c2545c9
+size 390

sortformer/v2-1/384_94MB/MelSpectrogram.mlmodelc/metadata.json

@@ -0,0 +1,76 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float32",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 3073 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 3073, 128]",
+        "name" : "melspectrogram_features",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Pad" : 1,
+      "Ios18.square" : 2,
+      "Ios18.conv" : 2,
+      "Ios18.sub" : 1,
+      "Ios18.expandDims" : 5,
+      "Ios18.matmul" : 1,
+      "Ios18.log" : 1,
+      "Ios18.concat" : 1,
+      "Ios18.add" : 2,
+      "Ios18.sliceByIndex" : 3,
+      "Ios18.cast" : 2,
+      "Ios18.transpose" : 1,
+      "Ios18.squeeze" : 2,
+      "Ios18.reshape" : 2,
+      "Identity" : 1,
+      "Ios18.mul" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-12",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 491520)",
+        "shortDescription" : "",
+        "shape" : "[491520]",
+        "name" : "audio",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "MelSpectrogram",
+    "method" : "predict"
+  }
+]

sortformer/v2-1/384_94MB/MelSpectrogram.mlmodelc/model.mil

@@ -0,0 +1,78 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3404.16.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
+{
+    func main<ios18>(tensor<fp16, [491520]> audio) {
+            string cast_0_dtype_0 = const()[name = string("cast_0_dtype_0"), val = string("fp32")];
+            tensor<fp32, [128, 257]> mel_filters = const()[name = string("mel_filters"), val = tensor<fp32, [128, 257]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
+            tensor<int32, [1]> var_6_begin_0 = const()[name = string("op_6_begin_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1]> var_6_end_0 = const()[name = string("op_6_end_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1]> var_6_end_mask_0 = const()[name = string("op_6_end_mask_0"), val = tensor<bool, [1]>([false])];
+            tensor<bool, [1]> var_6_squeeze_mask_0 = const()[name = string("op_6_squeeze_mask_0"), val = tensor<bool, [1]>([true])];
+            tensor<fp32, [491520]> cast_0 = cast(dtype = cast_0_dtype_0, x = audio)[name = string("cast_6")];
+            fp32 var_6 = slice_by_index(begin = var_6_begin_0, end = var_6_end_0, end_mask = var_6_end_mask_0, squeeze_mask = var_6_squeeze_mask_0, x = cast_0)[name = string("op_6")];
+            tensor<int32, [1]> var_8_axes_0 = const()[name = string("op_8_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [1]> var_8 = expand_dims(axes = var_8_axes_0, x = var_6)[name = string("op_8")];
+            tensor<int32, [1]> var_13_begin_0 = const()[name = string("op_13_begin_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_13_end_0 = const()[name = string("op_13_end_0"), val = tensor<int32, [1]>([491520])];
+            tensor<bool, [1]> var_13_end_mask_0 = const()[name = string("op_13_end_mask_0"), val = tensor<bool, [1]>([true])];
+            tensor<fp32, [491519]> var_13 = slice_by_index(begin = var_13_begin_0, end = var_13_end_0, end_mask = var_13_end_mask_0, x = cast_0)[name = string("op_13")];
+            tensor<int32, [1]> var_18_begin_0 = const()[name = string("op_18_begin_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1]> var_18_end_0 = const()[name = string("op_18_end_0"), val = tensor<int32, [1]>([491519])];
+            tensor<bool, [1]> var_18_end_mask_0 = const()[name = string("op_18_end_mask_0"), val = tensor<bool, [1]>([false])];
+            tensor<fp32, [491519]> var_18 = slice_by_index(begin = var_18_begin_0, end = var_18_end_0, end_mask = var_18_end_mask_0, x = cast_0)[name = string("op_18")];
+            fp32 var_19 = const()[name = string("op_19"), val = fp32(0x1.f0a3d8p-1)];
+            tensor<fp32, [491519]> var_20 = mul(x = var_18, y = var_19)[name = string("op_20")];
+            tensor<fp32, [491519]> var_22 = sub(x = var_13, y = var_20)[name = string("op_22")];
+            int32 var_24 = const()[name = string("op_24"), val = int32(0)];
+            bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)];
+            tensor<fp32, [491520]> input_1 = concat(axis = var_24, interleave = input_1_interleave_0, values = (var_8, var_22))[name = string("input_1")];
+            tensor<int32, [3]> var_32 = const()[name = string("op_32"), val = tensor<int32, [3]>([1, 1, 491520])];
+            tensor<fp32, [1, 1, 491520]> input_3 = reshape(shape = var_32, x = input_1)[name = string("input_3")];
+            fp32 const_1 = const()[name = string("const_1"), val = fp32(0x0p+0)];
+            tensor<int32, [6]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 256, 256])];
+            string input_5_mode_0 = const()[name = string("input_5_mode_0"), val = string("constant")];
+            tensor<fp32, [1, 1, 492032]> input_5 = pad(constant_val = const_1, mode = input_5_mode_0, pad = input_5_pad_0, x = input_3)[name = string("input_5")];
+            tensor<int32, [1]> var_44 = const()[name = string("op_44"), val = tensor<int32, [1]>([492032])];
+            tensor<fp32, [492032]> input = reshape(shape = var_44, x = input_5)[name = string("input")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [1, 492032]> expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = input)[name = string("expand_dims_0")];
+            tensor<fp32, [257, 1, 512]> expand_dims_1 = const()[name = string("expand_dims_1"), val = tensor<fp32, [257, 1, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131712)))];
+            tensor<fp32, [257, 1, 512]> expand_dims_2 = const()[name = string("expand_dims_2"), val = tensor<fp32, [257, 1, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(658112)))];
+            tensor<int32, [1]> expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor<int32, [1]>([160])];
+            tensor<int32, [1]> expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, 492032]> expand_dims_4 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0)[name = string("expand_dims_4")];
+            string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)];
+            tensor<fp32, [1, 257, 3073]> conv_0 = 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, x = expand_dims_4)[name = string("conv_0")];
+            string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)];
+            tensor<fp32, [1, 257, 3073]> conv_1 = 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, x = expand_dims_4)[name = string("conv_1")];
+            tensor<int32, [1]> squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [257, 3073]> squeeze_0 = squeeze(axes = squeeze_0_axes_0, x = conv_0)[name = string("squeeze_0")];
+            tensor<int32, [1]> squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [257, 3073]> squeeze_1 = squeeze(axes = squeeze_1_axes_0, x = conv_1)[name = string("squeeze_1")];
+            tensor<fp32, [257, 3073]> square_0 = square(x = squeeze_0)[name = string("square_0")];
+            tensor<fp32, [257, 3073]> square_1 = square(x = squeeze_1)[name = string("square_1")];
+            tensor<fp32, [257, 3073]> add_1 = add(x = square_0, y = square_1)[name = string("add_1")];
+            tensor<fp32, [257, 3073]> magnitudes = identity(x = add_1)[name = string("magnitudes")];
+            bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)];
+            bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)];
+            tensor<fp32, [128, 3073]> mel_spec_1 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters, y = magnitudes)[name = string("mel_spec_1")];
+            fp32 var_58 = const()[name = string("op_58"), val = fp32(0x1p-24)];
+            tensor<fp32, [128, 3073]> mel_spec_3 = add(x = mel_spec_1, y = var_58)[name = string("mel_spec_3")];
+            fp32 mel_spec_epsilon_0 = const()[name = string("mel_spec_epsilon_0"), val = fp32(0x1p-149)];
+            tensor<fp32, [128, 3073]> mel_spec = log(epsilon = mel_spec_epsilon_0, x = mel_spec_3)[name = string("mel_spec")];
+            tensor<int32, [2]> var_61_perm_0 = const()[name = string("op_61_perm_0"), val = tensor<int32, [2]>([1, 0])];
+            tensor<int32, [1]> var_63_axes_0 = const()[name = string("op_63_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [3073, 128]> var_61 = transpose(perm = var_61_perm_0, x = mel_spec)[name = string("transpose_0")];
+            tensor<fp32, [1, 3073, 128]> var_63 = expand_dims(axes = var_63_axes_0, x = var_61)[name = string("op_63")];
+            tensor<int32, [1]> var_65_axes_0 = const()[name = string("op_65_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, 3073, 128]> var_65 = expand_dims(axes = var_65_axes_0, x = var_63)[name = string("op_65")];
+            string cast_4_dtype_0 = const()[name = string("cast_4_dtype_0"), val = string("fp16")];
+            tensor<fp16, [1, 1, 3073, 128]> melspectrogram_features = cast(dtype = cast_4_dtype_0, x = var_65)[name = string("cast_5")];
+        } -> (melspectrogram_features);
+}

sortformer/v2-1/384_94MB/MelSpectrogram.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1570ad9beb3077cf61314606d8e4c30fc1d7b971e3bb73ee9444467faa9dd671
+size 1184512

sortformer/v2-1/384_94MB/SortformerFullEncoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8404f0c6e336a3a535d067e63c398d62dacaa31bd0f9b8bfb064414dcd70f597
+size 243

sortformer/v2-1/384_94MB/SortformerFullEncoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0306109bf01fe682b6d1c6fb6d0c24e1a06e7aee76545a51bc0b381d7ab5262
+size 609

sortformer/v2-1/384_94MB/SortformerFullEncoder.mlmodelc/metadata.json

@@ -0,0 +1,127 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Mixed (Float16, Palettized (6 bits))",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 4 × 1 × 384)",
+        "shortDescription" : "",
+        "shape" : "[1, 4, 1, 384]",
+        "name" : "raw_speaker_preds",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 4 × 1 × 384)",
+        "shortDescription" : "",
+        "shape" : "[1, 4, 1, 384]",
+        "name" : "speaker_sigmoids",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 9,
+    "mlProgramOperationTypeHistogram" : {
+      "Pad" : 17,
+      "Ios18.constexprLutToDense" : 315,
+      "Ios18.conv" : 315,
+      "Ios18.sub" : 1,
+      "Ios18.matmul" : 87,
+      "Ios18.batchNorm" : 121,
+      "Ios18.expandDims" : 4,
+      "Ios18.relu" : 20,
+      "Ios18.sigmoid" : 18,
+      "Ios18.add" : 190,
+      "Ios18.silu" : 51,
+      "Ios18.softmax" : 35,
+      "Ios18.sliceByIndex" : 34,
+      "Ios18.layerNorm" : 121,
+      "Ios18.reshape" : 208,
+      "Split" : 17,
+      "Ios18.mul" : 105
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "15.0",
+      "tvOS" : "18.0",
+      "visionOS" : "2.0",
+      "watchOS" : "11.0",
+      "iOS" : "18.0",
+      "macCatalyst" : "18.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2026-02-12",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "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 × 512 × 1 × 384)",
+        "shortDescription" : "",
+        "shape" : "[1, 512, 1, 384]",
+        "name" : "downsampled_melspectrogram_features",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 384)",
+        "shortDescription" : "",
+        "shape" : "[1, 384]",
+        "name" : "conformer_encoder_padding_mask",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 384 × 384)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 384, 384]",
+        "name" : "conformer_encoder_qk_mask",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 384)",
+        "shortDescription" : "",
+        "shape" : "[1, 384]",
+        "name" : "transformer_encoder_mask",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1, 1]",
+        "name" : "input_1",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "SortformerFullEncoder_6_bit",
+    "method" : "predict"
+  }
+]

sortformer/v2-1/384_94MB/SortformerFullEncoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:381c70fa727f11b773b801654404288dcb45b6033d35cde4ed5b1ed4776b189a
+size 87351040