Add CIF alphas model (host does only integrate-and-fire)

ParaformerCifAlphas.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eab07ffe1555c52bf453304c7440181d05cb813c507f7d2b9dc690f893daee7f
+size 243

ParaformerCifAlphas.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fdc3a31784fdd4ce76d8f6662cc1d5266a4bcb9f8ebff1f7364e4af9b9bfb1b
+size 358

ParaformerCifAlphas.mlmodelc/model.mil

@@ -0,0 +1,32 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
+{
+    func main<ios17>(tensor<fp32, [1, ?, 512]> enc_out) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>>>((("DefaultShapes", {{"enc_out", [1, 1800, 512]}}), ("EnumeratedShapes", {{"enc_out_1_1_1_1024_512_", {{"enc_out", [1, 1024, 512]}}}, {"enc_out_1_1_1_128_512_", {{"enc_out", [1, 128, 512]}}}, {"enc_out_1_1_1_1800_512_", {{"enc_out", [1, 1800, 512]}}}, {"enc_out_1_1_1_256_512_", {{"enc_out", [1, 256, 512]}}}, {"enc_out_1_1_1_512_512_", {{"enc_out", [1, 512, 512]}}}})))] {
+            tensor<int32, [3]> input_1_perm_0 = const()[name = tensor<string, []>("input_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> enc_out_to_fp16_dtype_0 = const()[name = tensor<string, []>("enc_out_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<int32, [6]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 1])];
+            tensor<string, []> input_3_mode_0 = const()[name = tensor<string, []>("input_3_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, ?, 512]> enc_out_to_fp16 = cast(dtype = enc_out_to_fp16_dtype_0, x = enc_out)[name = tensor<string, []>("cast_1")];
+            tensor<fp16, [1, 512, ?]> input_1_cast_fp16 = transpose(perm = input_1_perm_0, x = enc_out_to_fp16)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 512, ?]> input_3_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<string, []> var_24_pad_type_0 = const()[name = tensor<string, []>("op_24_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_24_strides_0 = const()[name = tensor<string, []>("op_24_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_24_pad_0 = const()[name = tensor<string, []>("op_24_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_24_dilations_0 = const()[name = tensor<string, []>("op_24_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_24_groups_0 = const()[name = tensor<string, []>("op_24_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [512, 512, 3]> p_cif_conv1d_weight_to_fp16 = const()[name = tensor<string, []>("p_cif_conv1d_weight_to_fp16"), val = tensor<fp16, [512, 512, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [512]> p_cif_conv1d_bias_to_fp16 = const()[name = tensor<string, []>("p_cif_conv1d_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572992)))];
+            tensor<fp16, [1, 512, ?]> var_24_cast_fp16 = conv(bias = p_cif_conv1d_bias_to_fp16, dilations = var_24_dilations_0, groups = var_24_groups_0, pad = var_24_pad_0, pad_type = var_24_pad_type_0, strides = var_24_strides_0, weight = p_cif_conv1d_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("op_24_cast_fp16")];
+            tensor<fp16, [1, 512, ?]> var_25_cast_fp16 = relu(x = var_24_cast_fp16)[name = tensor<string, []>("op_25_cast_fp16")];
+            tensor<int32, [3]> input_5_perm_0 = const()[name = tensor<string, []>("input_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 512]> p_cif_output_weight_to_fp16 = const()[name = tensor<string, []>("p_cif_output_weight_to_fp16"), val = tensor<fp16, [1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574080)))];
+            tensor<fp16, [1]> p_cif_output_bias_to_fp16 = const()[name = tensor<string, []>("p_cif_output_bias_to_fp16"), val = tensor<fp16, [1]>([-0x1.c3cp-4])];
+            tensor<fp16, [1, ?, 512]> input_5_cast_fp16 = transpose(perm = input_5_perm_0, x = var_25_cast_fp16)[name = tensor<string, []>("transpose_0")];
+            tensor<fp16, [1, ?, 1]> linear_0_cast_fp16 = linear(bias = p_cif_output_bias_to_fp16, weight = p_cif_output_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<fp16, [1, ?, 1]> a_1_cast_fp16 = sigmoid(x = linear_0_cast_fp16)[name = tensor<string, []>("a_1_cast_fp16")];
+            tensor<fp16, [1, ?, 1]> a_cast_fp16 = relu(x = a_1_cast_fp16)[name = tensor<string, []>("a_cast_fp16")];
+            tensor<int32, [1]> var_40_axes_0 = const()[name = tensor<string, []>("op_40_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, ?]> alphas = squeeze(axes = var_40_axes_0, x = a_cast_fp16)[name = tensor<string, []>("op_40_cast_fp16")];
+        } -> (alphas);
+}

ParaformerCifAlphas.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca67890e3fd418ca1f3a227001ba4f1300d5523b8a49470bf3791eaba836bbee
+size 1575168