ios17/multilingual/560ms (iOS17 deployment target, same recipe)

ios17/multilingual/560ms/decoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdb14a08e42b4806a2d1505501586be71e4f04ca9256c719544fd7ed6937e509
+size 243

ios17/multilingual/560ms/decoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:754cd5f784e66cd0361f13c211141f94d15f4269a354c1531bb98c5722b88251
+size 433

ios17/multilingual/560ms/decoder.mlmodelc/model.mil

@@ -0,0 +1,64 @@
+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, [2, 1, 640]> c_in, tensor<fp32, [2, 1, 640]> h_in, tensor<int32, [1, 1]> token, tensor<int32, [1]> token_length) {
+            tensor<int32, []> y_axis_0 = const()[name = tensor<string, []>("y_axis_0"), val = tensor<int32, []>(0)];
+            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, [13088, 640]> module_prediction_embed_weight_to_fp16 = const()[name = tensor<string, []>("module_prediction_embed_weight_to_fp16"), val = tensor<fp16, [13088, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<string, []> token_to_int16_dtype_0 = const()[name = tensor<string, []>("token_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1]> token_to_int16 = cast(dtype = token_to_int16_dtype_0, x = token)[name = tensor<string, []>("cast_8")];
+            tensor<fp16, [1, 1, 640]> y_cast_fp16_cast_uint16 = gather(axis = y_axis_0, batch_dims = y_batch_dims_0, indices = token_to_int16, validate_indices = y_validate_indices_0, x = module_prediction_embed_weight_to_fp16)[name = tensor<string, []>("y_cast_fp16_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_7")];
+            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_6")];
+            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, []>(16752768)))];
+            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, []>(20029632)))];
+            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, []>(23306496)))];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = y_cast_fp16_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, []>(23311680)))];
+            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, []>(26588544)))];
+            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, []>(29865408)))];
+            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_out = cast(dtype = transpose_0_cast_fp16_to_fp32_dtype_0, x = transpose_0_cast_fp16)[name = tensor<string, []>("cast_3")];
+            tensor<fp32, [2, 1, 640]> c_out = cast(dtype = obj_cast_fp16_to_fp32_dtype_0, x = obj_cast_fp16)[name = tensor<string, []>("cast_4")];
+            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_5")];
+            tensor<int32, [1]> token_length_tmp = identity(x = token_length)[name = tensor<string, []>("token_length_tmp")];
+        } -> (decoder_out, h_out, c_out);
+}

ios17/multilingual/560ms/decoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dcdeccd4ccf46e2675224f9f030d46c1a89e2bda4abb316e901e1a21f1597f8f
+size 29870592

ios17/multilingual/560ms/decoder_joint.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a8e98a54ed1f16c3d5125816a002b991e167d620beb8fcc557f26d9a1c092f8
+size 243

ios17/multilingual/560ms/decoder_joint.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddd2d71cd9d13cc7b1d6f7632ab99f482b5f3f07a726bbc0673802c9dd6dcf71
+size 454

ios17/multilingual/560ms/decoder_joint.mlmodelc/model.mil

@@ -0,0 +1,83 @@
+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, [2, 1, 640]> c_in, tensor<fp32, [1, 1024, 1]> encoder, tensor<fp32, [2, 1, 640]> h_in, tensor<int32, [1, 1]> token, tensor<int32, [1]> token_length) {
+            tensor<int32, []> y_axis_0 = const()[name = tensor<string, []>("y_axis_0"), val = tensor<int32, []>(0)];
+            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, [13088, 640]> decoder_module_prediction_embed_weight_to_fp16 = const()[name = tensor<string, []>("decoder_module_prediction_embed_weight_to_fp16"), val = tensor<fp16, [13088, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<string, []> token_to_int16_dtype_0 = const()[name = tensor<string, []>("token_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<int16, [1, 1]> token_to_int16 = cast(dtype = token_to_int16_dtype_0, x = token)[name = tensor<string, []>("cast_9")];
+            tensor<fp16, [1, 1, 640]> y_cast_fp16_cast_uint16 = gather(axis = y_axis_0, batch_dims = y_batch_dims_0, indices = token_to_int16, validate_indices = y_validate_indices_0, x = decoder_module_prediction_embed_weight_to_fp16)[name = tensor<string, []>("y_cast_fp16_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_8")];
+            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_7")];
+            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_5_lstm_layer_0_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("input_5_lstm_layer_0_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_5_lstm_layer_0_lstm_h0_squeeze_cast_fp16 = squeeze(axes = input_5_lstm_layer_0_lstm_h0_squeeze_axes_0, x = split_0_cast_fp16_0)[name = tensor<string, []>("input_5_lstm_layer_0_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> input_5_lstm_layer_0_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("input_5_lstm_layer_0_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_5_lstm_layer_0_lstm_c0_squeeze_cast_fp16 = squeeze(axes = input_5_lstm_layer_0_lstm_c0_squeeze_axes_0, x = split_1_cast_fp16_0)[name = tensor<string, []>("input_5_lstm_layer_0_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> input_5_lstm_layer_0_direction_0 = const()[name = tensor<string, []>("input_5_lstm_layer_0_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> input_5_lstm_layer_0_output_sequence_0 = const()[name = tensor<string, []>("input_5_lstm_layer_0_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_5_lstm_layer_0_recurrent_activation_0 = const()[name = tensor<string, []>("input_5_lstm_layer_0_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_5_lstm_layer_0_cell_activation_0 = const()[name = tensor<string, []>("input_5_lstm_layer_0_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_5_lstm_layer_0_activation_0 = const()[name = tensor<string, []>("input_5_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, []>(16752768)))];
+            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, []>(20029632)))];
+            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, []>(23306496)))];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = y_cast_fp16_cast_uint16)[name = tensor<string, []>("transpose_4")];
+            tensor<fp16, [1, 1, 640]> input_5_lstm_layer_0_cast_fp16_0, tensor<fp16, [1, 640]> input_5_lstm_layer_0_cast_fp16_1, tensor<fp16, [1, 640]> input_5_lstm_layer_0_cast_fp16_2 = lstm(activation = input_5_lstm_layer_0_activation_0, bias = concat_0_to_fp16, cell_activation = input_5_lstm_layer_0_cell_activation_0, direction = input_5_lstm_layer_0_direction_0, initial_c = input_5_lstm_layer_0_lstm_c0_squeeze_cast_fp16, initial_h = input_5_lstm_layer_0_lstm_h0_squeeze_cast_fp16, output_sequence = input_5_lstm_layer_0_output_sequence_0, recurrent_activation = input_5_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_5_lstm_layer_0_cast_fp16")];
+            tensor<int32, [1]> input_5_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("input_5_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_5_lstm_h0_squeeze_cast_fp16 = squeeze(axes = input_5_lstm_h0_squeeze_axes_0, x = split_0_cast_fp16_1)[name = tensor<string, []>("input_5_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> input_5_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("input_5_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 640]> input_5_lstm_c0_squeeze_cast_fp16 = squeeze(axes = input_5_lstm_c0_squeeze_axes_0, x = split_1_cast_fp16_1)[name = tensor<string, []>("input_5_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> input_5_direction_0 = const()[name = tensor<string, []>("input_5_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> input_5_output_sequence_0 = const()[name = tensor<string, []>("input_5_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> input_5_recurrent_activation_0 = const()[name = tensor<string, []>("input_5_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> input_5_cell_activation_0 = const()[name = tensor<string, []>("input_5_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> input_5_activation_0 = const()[name = tensor<string, []>("input_5_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, []>(23311680)))];
+            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, []>(26588544)))];
+            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, []>(29865408)))];
+            tensor<fp16, [1, 1, 640]> input_5_cast_fp16_0, tensor<fp16, [1, 640]> input_5_cast_fp16_1, tensor<fp16, [1, 640]> input_5_cast_fp16_2 = lstm(activation = input_5_activation_0, bias = concat_3_to_fp16, cell_activation = input_5_cell_activation_0, direction = input_5_direction_0, initial_c = input_5_lstm_c0_squeeze_cast_fp16, initial_h = input_5_lstm_h0_squeeze_cast_fp16, output_sequence = input_5_output_sequence_0, recurrent_activation = input_5_recurrent_activation_0, weight_hh = concat_5_to_fp16, weight_ih = concat_4_to_fp16, x = input_5_lstm_layer_0_cast_fp16_0)[name = tensor<string, []>("input_5_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_5_lstm_layer_0_cast_fp16_1, input_5_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_5_lstm_layer_0_cast_fp16_2, input_5_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_1_perm_0 = const()[name = tensor<string, []>("transpose_1_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
+            tensor<int32, [3]> input_7_perm_0 = const()[name = tensor<string, []>("input_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> encoder_to_fp16_dtype_0 = const()[name = tensor<string, []>("encoder_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, []>(29870592)))];
+            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, []>(31181376)))];
+            tensor<fp16, [1, 1024, 1]> encoder_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = tensor<string, []>("cast_4")];
+            tensor<fp16, [1, 1, 1024]> input_7_cast_fp16 = transpose(perm = input_7_perm_0, x = encoder_to_fp16)[name = tensor<string, []>("transpose_2")];
+            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_7_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, []>(31182720)))];
+            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, []>(32001984)))];
+            tensor<fp16, [1, 1, 640]> transpose_1_cast_fp16 = transpose(perm = transpose_1_perm_0, x = input_5_cast_fp16_0)[name = tensor<string, []>("transpose_3")];
+            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 = transpose_1_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
+            tensor<int32, [1]> var_79_axes_0 = const()[name = tensor<string, []>("op_79_axes_0"), val = tensor<int32, [1]>([2])];
+            tensor<fp16, [1, 1, 1, 640]> var_79_cast_fp16 = expand_dims(axes = var_79_axes_0, x = linear_0_cast_fp16)[name = tensor<string, []>("op_79_cast_fp16")];
+            tensor<int32, [1]> var_80_axes_0 = const()[name = tensor<string, []>("op_80_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, 1, 640]> var_80_cast_fp16 = expand_dims(axes = var_80_axes_0, x = linear_1_cast_fp16)[name = tensor<string, []>("op_80_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_11_cast_fp16 = add(x = var_79_cast_fp16, y = var_80_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_13_cast_fp16 = relu(x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
+            tensor<fp16, [13088, 640]> joint_module_joint_net_2_weight_to_fp16 = const()[name = tensor<string, []>("joint_module_joint_net_2_weight_to_fp16"), val = tensor<fp16, [13088, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32003328)))];
+            tensor<fp16, [13088]> joint_module_joint_net_2_bias_to_fp16 = const()[name = tensor<string, []>("joint_module_joint_net_2_bias_to_fp16"), val = tensor<fp16, [13088]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48756032)))];
+            tensor<fp16, [1, 1, 1, 13088]> 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_13_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<string, []> linear_2_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_2_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 1, 1, 13088]> logits = cast(dtype = linear_2_cast_fp16_to_fp32_dtype_0, x = linear_2_cast_fp16)[name = tensor<string, []>("cast_3")];
+            tensor<fp32, [2, 1, 640]> c_out = cast(dtype = obj_cast_fp16_to_fp32_dtype_0, x = obj_cast_fp16)[name = tensor<string, []>("cast_5")];
+            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_6")];
+            tensor<int32, [1]> token_length_tmp = identity(x = token_length)[name = tensor<string, []>("token_length_tmp")];
+        } -> (logits, h_out, c_out);
+}

ios17/multilingual/560ms/decoder_joint.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01f21eb747fbc53bd0ed7efebea1bf0aa655ebf2816f21d0bb6554c9b7fcfc0b
+size 48782272

ios17/multilingual/560ms/encoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3409887ed9eadc8689bd4c64754d95e4119152dc0b9ee1650664d2a6a689c783
+size 243

ios17/multilingual/560ms/encoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:269d7188428298be3d0b2803c0e42bf6d44f7b4a9371030ae690beac443afbfa
+size 572

ios17/multilingual/560ms/encoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e994440ed6c936ac375d375eaab16eb445881ff2d9bc62384de991b411e234e
+size 564307200

ios17/multilingual/560ms/joint.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1a90a7d5f8b86f564a42ab45b42a43eb0bbce6682176916bd94591b28cca447
+size 243

ios17/multilingual/560ms/joint.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94b7eac58dc52a40bff8965c04d941357fed87c77265f61a0288a289ad9e9dfb
+size 341

ios17/multilingual/560ms/joint.mlmodelc/model.mil

@@ -0,0 +1,31 @@
+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, 640, 1]> decoder, tensor<fp32, [1, 1024, 1]> encoder) {
+            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_to_fp16_dtype_0 = const()[name = tensor<string, []>("encoder_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_to_fp16_dtype_0 = const()[name = tensor<string, []>("decoder_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [640, 1024]> module_enc_weight_to_fp16 = const()[name = tensor<string, []>("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]> module_enc_bias_to_fp16 = const()[name = tensor<string, []>("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_to_fp16 = cast(dtype = encoder_to_fp16_dtype_0, x = encoder)[name = tensor<string, []>("cast_2")];
+            tensor<fp16, [1, 1, 1024]> input_1_cast_fp16 = transpose(perm = input_1_perm_0, x = encoder_to_fp16)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 1, 640]> linear_0_cast_fp16 = linear(bias = module_enc_bias_to_fp16, weight = module_enc_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<fp16, [640, 640]> module_pred_weight_to_fp16 = const()[name = tensor<string, []>("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]> module_pred_bias_to_fp16 = const()[name = tensor<string, []>("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_to_fp16 = cast(dtype = decoder_to_fp16_dtype_0, x = decoder)[name = tensor<string, []>("cast_1")];
+            tensor<fp16, [1, 1, 640]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = decoder_to_fp16)[name = tensor<string, []>("transpose_0")];
+            tensor<fp16, [1, 1, 640]> linear_1_cast_fp16 = linear(bias = module_pred_bias_to_fp16, weight = 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_25_axes_0 = const()[name = tensor<string, []>("op_25_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, 1, 640]> var_25_cast_fp16 = expand_dims(axes = var_25_axes_0, x = linear_1_cast_fp16)[name = tensor<string, []>("op_25_cast_fp16")];
+            tensor<fp16, [1, 1, 1, 640]> input_5_cast_fp16 = add(x = var_23_cast_fp16, y = var_25_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, [13088, 640]> module_joint_net_2_weight_to_fp16 = const()[name = tensor<string, []>("module_joint_net_2_weight_to_fp16"), val = tensor<fp16, [13088, 640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2132800)))];
+            tensor<fp16, [13088]> module_joint_net_2_bias_to_fp16 = const()[name = tensor<string, []>("module_joint_net_2_bias_to_fp16"), val = tensor<fp16, [13088]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18885504)))];
+            tensor<fp16, [1, 1, 1, 13088]> linear_2_cast_fp16 = linear(bias = module_joint_net_2_bias_to_fp16, weight = module_joint_net_2_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<string, []> linear_2_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_2_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 1, 1, 13088]> logits = cast(dtype = linear_2_cast_fp16_to_fp32_dtype_0, x = linear_2_cast_fp16)[name = tensor<string, []>("cast_0")];
+        } -> (logits);
+}

ios17/multilingual/560ms/joint.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0ef0a3a6598f962d2aad598dc6850e4428874033419817121e11f1fff4a9cfe
+size 18911744

ios17/multilingual/560ms/metadata.json

@@ -0,0 +1,196 @@
+{
+  "model": "nvidia/nemotron-asr-streaming-multilingual-0.6b",
+  "model_class": "nemo.collections.asr.models.rnnt_bpe_models_prompt.EncDecRNNTBPEModelWithPrompt",
+  "sample_rate": 16000,
+  "mel_features": 128,
+  "chunk_mel_frames": 56,
+  "pre_encode_cache": 9,
+  "total_mel_frames": 65,
+  "att_context_size": [
+    42,
+    13
+  ],
+  "vocab_size": 13087,
+  "blank_idx": 13087,
+  "cache_channel_shape": [
+    1,
+    24,
+    42,
+    1024
+  ],
+  "cache_time_shape": [
+    1,
+    24,
+    1024,
+    8
+  ],
+  "decoder_hidden": 640,
+  "decoder_layers": 2,
+  "encoder_dim": 1024,
+  "num_prompts": 128,
+  "prompt_dictionary": {
+    "en-US": 0,
+    "en": 0,
+    "en-GB": 1,
+    "enGB": 1,
+    "es-ES": 2,
+    "esES": 2,
+    "es-US": 3,
+    "es": 3,
+    "zh-CN": 4,
+    "zh-ZH": 4,
+    "zh-TW": 5,
+    "hi-IN": 6,
+    "hi": 6,
+    "hi-HI": 6,
+    "ar-AR": 7,
+    "ar": 7,
+    "fr-FR": 8,
+    "fr": 8,
+    "de-DE": 9,
+    "de": 9,
+    "ja-JP": 10,
+    "ja-JA": 10,
+    "ru-RU": 11,
+    "ru": 11,
+    "pt-BR": 12,
+    "pt-PT": 13,
+    "pt": 13,
+    "ko-KR": 14,
+    "ko": 14,
+    "ko-KO": 14,
+    "it-IT": 15,
+    "it": 15,
+    "nl-NL": 16,
+    "nl": 16,
+    "pl-PL": 17,
+    "pl": 17,
+    "tr-TR": 18,
+    "tr": 18,
+    "uk-UA": 19,
+    "uk": 19,
+    "ro-RO": 20,
+    "ro": 20,
+    "el-GR": 21,
+    "el": 21,
+    "cs-CZ": 22,
+    "cs": 22,
+    "hu-HU": 23,
+    "hu": 23,
+    "sv-SE": 24,
+    "sv": 24,
+    "da-DK": 25,
+    "da": 25,
+    "fi-FI": 26,
+    "fi": 26,
+    "no-NO": 27,
+    "no": 27,
+    "nb-NO": 103,
+    "nb": 103,
+    "nn-NO": 104,
+    "nn": 104,
+    "sk-SK": 28,
+    "sk": 28,
+    "hr-HR": 29,
+    "hr": 29,
+    "bg-BG": 30,
+    "bg": 30,
+    "lt-LT": 31,
+    "lt": 31,
+    "et-EE": 60,
+    "et": 60,
+    "lv-LV": 61,
+    "lv": 61,
+    "sl-SI": 62,
+    "sl": 62,
+    "th-TH": 32,
+    "vi-VN": 33,
+    "id-ID": 34,
+    "ms-MY": 35,
+    "bn-IN": 36,
+    "ur-PK": 37,
+    "fa-IR": 38,
+    "ta-IN": 39,
+    "te-IN": 40,
+    "mr-IN": 41,
+    "gu-IN": 42,
+    "kn-IN": 43,
+    "ml-IN": 44,
+    "si-LK": 45,
+    "ne-NP": 46,
+    "km-KH": 47,
+    "sw-KE": 48,
+    "am-ET": 49,
+    "ha-NG": 50,
+    "zu-ZA": 51,
+    "yo-NG": 52,
+    "ig-NG": 53,
+    "af-ZA": 54,
+    "rw-RW": 55,
+    "so-SO": 56,
+    "ny-MW": 57,
+    "ln-CD": 58,
+    "or-KE": 59,
+    "he-IL": 64,
+    "ku-TR": 65,
+    "az-AZ": 66,
+    "ka-GE": 67,
+    "hy-AM": 68,
+    "uz-UZ": 69,
+    "tg-TJ": 70,
+    "ky-KG": 71,
+    "qu-PE": 80,
+    "ay-BO": 81,
+    "gn-PY": 82,
+    "nah-MX": 83,
+    "mi-NZ": 96,
+    "haw-US": 97,
+    "sm-WS": 98,
+    "to-TO": 99,
+    "fr-CA": 100,
+    "mt-MT": 102,
+    "auto": 101
+  },
+  "default_prompt_id": 101,
+  "lang_tag_token_ids": [
+    1,
+    256,
+    397,
+    518,
+    673,
+    814,
+    907,
+    993,
+    1125,
+    1232,
+    1279,
+    1383,
+    1455,
+    1603,
+    1724,
+    1841,
+    1929,
+    2021,
+    2124,
+    2205,
+    2322,
+    2440,
+    2529,
+    2809,
+    2947,
+    2986,
+    3051,
+    3064,
+    3134,
+    3247,
+    3446,
+    7489,
+    9532,
+    9544,
+    9596,
+    9695,
+    9815,
+    9847,
+    12944
+  ]
+}

ios17/multilingual/560ms/preprocessor.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d628a4c3b5a7f15375b09ae18a50d608ac0cce12de0e6d0749d7df2e51101e1
+size 243

ios17/multilingual/560ms/preprocessor.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d1aa8c8e7e283e4944af4b0b701db760ed99ef14919d3f989c599b9f63335a2
+size 371

ios17/multilingual/560ms/preprocessor.mlmodelc/model.mil

@@ -0,0 +1,122 @@
+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, ?]> audio, tensor<int32, [1]> audio_length) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"audio", [1, 1]}}), ("RangeDims", {{"audio", [[1, 1], [1, 1280000]]}})))] {
+            tensor<int32, []> var_9 = const()[name = tensor<string, []>("op_9"), val = tensor<int32, []>(1)];
+            tensor<int32, []> var_10 = const()[name = tensor<string, []>("op_10"), val = tensor<int32, []>(160)];
+            tensor<int32, []> var_12 = const()[name = tensor<string, []>("op_12"), val = tensor<int32, []>(0)];
+            tensor<int32, []> var_33 = const()[name = tensor<string, []>("op_33"), val = tensor<int32, []>(512)];
+            tensor<int32, [1]> var_34 = add(x = audio_length, y = var_33)[name = tensor<string, []>("op_34")];
+            tensor<int32, []> var_35 = const()[name = tensor<string, []>("op_35"), val = tensor<int32, []>(512)];
+            tensor<int32, [1]> var_36 = sub(x = var_34, y = var_35)[name = tensor<string, []>("op_36")];
+            tensor<int32, [1]> floor_div_0 = floor_div(x = var_36, y = var_10)[name = tensor<string, []>("floor_div_0")];
+            tensor<bool, [1]> var_39 = equal(x = audio_length, y = var_12)[name = tensor<string, []>("op_39")];
+            tensor<int32, [1]> var_40 = const()[name = tensor<string, []>("op_40"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1]> mel_length = select(a = var_40, b = floor_div_0, cond = var_39)[name = tensor<string, []>("seq_len")];
+            tensor<string, []> audio_to_fp16_dtype_0 = const()[name = tensor<string, []>("audio_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, ?]> audio_to_fp16 = cast(dtype = audio_to_fp16_dtype_0, x = audio)[name = tensor<string, []>("cast_14")];
+            tensor<int32, [2]> var_42_shape_cast_fp16 = shape(x = audio_to_fp16)[name = tensor<string, []>("op_42_shape_cast_fp16")];
+            tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> gather_0_validate_indices_0 = const()[name = tensor<string, []>("gather_0_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_42_shape_cast_fp16_to_int16_dtype_0 = const()[name = tensor<string, []>("op_42_shape_cast_fp16_to_int16_dtype_0"), val = tensor<string, []>("int16")];
+            tensor<uint16, []> select_0_to_uint16 = const()[name = tensor<string, []>("select_0_to_uint16"), val = tensor<uint16, []>(1)];
+            tensor<int16, [2]> var_42_shape_cast_fp16_to_int16 = cast(dtype = var_42_shape_cast_fp16_to_int16_dtype_0, x = var_42_shape_cast_fp16)[name = tensor<string, []>("cast_13")];
+            tensor<int16, []> gather_0_cast_uint16 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = select_0_to_uint16, validate_indices = gather_0_validate_indices_0, x = var_42_shape_cast_fp16_to_int16)[name = tensor<string, []>("gather_0_cast_uint16")];
+            tensor<string, []> gather_0_cast_uint16_to_int32_dtype_0 = const()[name = tensor<string, []>("gather_0_cast_uint16_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> const_1 = const()[name = tensor<string, []>("const_1"), val = tensor<int32, []>(1)];
+            tensor<int32, []> gather_0_cast_uint16_to_int32 = cast(dtype = gather_0_cast_uint16_to_int32_dtype_0, x = gather_0_cast_uint16)[name = tensor<string, []>("cast_12")];
+            tensor<int32, [?]> var_43 = range_1d(end = gather_0_cast_uint16_to_int32, start = const_0, step = const_1)[name = tensor<string, []>("op_43")];
+            tensor<int32, [1]> var_44_axes_0 = const()[name = tensor<string, []>("op_44_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1, ?]> var_44 = expand_dims(axes = var_44_axes_0, x = var_43)[name = tensor<string, []>("op_44")];
+            tensor<int32, [1]> var_45_axes_0 = const()[name = tensor<string, []>("op_45_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1, 1]> var_45 = expand_dims(axes = var_45_axes_0, x = audio_length)[name = tensor<string, []>("op_45")];
+            tensor<bool, [1, ?]> timemask = less(x = var_44, y = var_45)[name = tensor<string, []>("timemask")];
+            tensor<int32, [2]> var_48_begin_0 = const()[name = tensor<string, []>("op_48_begin_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [2]> var_48_end_0 = const()[name = tensor<string, []>("op_48_end_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<bool, [2]> var_48_end_mask_0 = const()[name = tensor<string, []>("op_48_end_mask_0"), val = tensor<bool, [2]>([true, false])];
+            tensor<bool, [2]> var_48_squeeze_mask_0 = const()[name = tensor<string, []>("op_48_squeeze_mask_0"), val = tensor<bool, [2]>([false, true])];
+            tensor<fp16, [1]> var_48_cast_fp16 = slice_by_index(begin = var_48_begin_0, end = var_48_end_0, end_mask = var_48_end_mask_0, squeeze_mask = var_48_squeeze_mask_0, x = audio_to_fp16)[name = tensor<string, []>("op_48_cast_fp16")];
+            tensor<int32, [1]> var_49_axes_0 = const()[name = tensor<string, []>("op_49_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1]> var_49_cast_fp16 = expand_dims(axes = var_49_axes_0, x = var_48_cast_fp16)[name = tensor<string, []>("op_49_cast_fp16")];
+            tensor<int32, [2]> var_51_begin_0 = const()[name = tensor<string, []>("op_51_begin_0"), val = tensor<int32, [2]>([0, 1])];
+            tensor<int32, [2]> var_51_end_0 = const()[name = tensor<string, []>("op_51_end_0"), val = tensor<int32, [2]>([1, 0])];
+            tensor<bool, [2]> var_51_end_mask_0 = const()[name = tensor<string, []>("op_51_end_mask_0"), val = tensor<bool, [2]>([true, true])];
+            tensor<fp16, [1, ?]> var_51_cast_fp16 = slice_by_index(begin = var_51_begin_0, end = var_51_end_0, end_mask = var_51_end_mask_0, x = audio_to_fp16)[name = tensor<string, []>("op_51_cast_fp16")];
+            tensor<int32, [2]> var_53_begin_0 = const()[name = tensor<string, []>("op_53_begin_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [2]> var_53_end_0 = const()[name = tensor<string, []>("op_53_end_0"), val = tensor<int32, [2]>([1, -1])];
+            tensor<bool, [2]> var_53_end_mask_0 = const()[name = tensor<string, []>("op_53_end_mask_0"), val = tensor<bool, [2]>([true, false])];
+            tensor<fp16, [1, ?]> var_53_cast_fp16 = slice_by_index(begin = var_53_begin_0, end = var_53_end_0, end_mask = var_53_end_mask_0, x = audio_to_fp16)[name = tensor<string, []>("op_53_cast_fp16")];
+            tensor<fp16, []> var_54_to_fp16 = const()[name = tensor<string, []>("op_54_to_fp16"), val = tensor<fp16, []>(0x1.f0cp-1)];
+            tensor<fp16, [1, ?]> var_55_cast_fp16 = mul(x = var_53_cast_fp16, y = var_54_to_fp16)[name = tensor<string, []>("op_55_cast_fp16")];
+            tensor<fp16, [1, ?]> var_56_cast_fp16 = sub(x = var_51_cast_fp16, y = var_55_cast_fp16)[name = tensor<string, []>("op_56_cast_fp16")];
+            tensor<bool, []> x_3_interleave_0 = const()[name = tensor<string, []>("x_3_interleave_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, ?]> x_3_cast_fp16 = concat(axis = var_9, interleave = x_3_interleave_0, values = (var_49_cast_fp16, var_56_cast_fp16))[name = tensor<string, []>("x_3_cast_fp16")];
+            tensor<bool, [1, ?]> var_59 = logical_not(x = timemask)[name = tensor<string, []>("op_59")];
+            tensor<fp16, []> var_16_to_fp16 = const()[name = tensor<string, []>("op_16_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, ?]> input_1_cast_fp16 = select(a = var_16_to_fp16, b = x_3_cast_fp16, cond = var_59)[name = tensor<string, []>("input_1_cast_fp16")];
+            tensor<int32, [3]> concat_1x = const()[name = tensor<string, []>("concat_1x"), val = tensor<int32, [3]>([1, 1, -1])];
+            tensor<fp16, [1, 1, ?]> input_3_cast_fp16 = reshape(shape = concat_1x, x = input_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<int32, [6]> input_5_pad_0 = const()[name = tensor<string, []>("input_5_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 256, 256])];
+            tensor<string, []> input_5_mode_0 = const()[name = tensor<string, []>("input_5_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_3_to_fp16 = const()[name = tensor<string, []>("const_3_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1, ?]> input_5_cast_fp16 = pad(constant_val = const_3_to_fp16, mode = input_5_mode_0, pad = input_5_pad_0, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<int32, [2]> concat_2x = const()[name = tensor<string, []>("concat_2x"), val = tensor<int32, [2]>([1, -1])];
+            tensor<fp16, [1, ?]> input_cast_fp16 = reshape(shape = concat_2x, x = input_5_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<int32, [1]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1]>([160])];
+            tensor<int32, [1]> expand_dims_4_axes_0 = const()[name = tensor<string, []>("expand_dims_4_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 1, ?]> expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = input_cast_fp16)[name = tensor<string, []>("expand_dims_4_cast_fp16")];
+            tensor<string, []> conv_0_pad_type_0 = const()[name = tensor<string, []>("conv_0_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_0_pad_0 = const()[name = tensor<string, []>("conv_0_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_0_dilations_0 = const()[name = tensor<string, []>("conv_0_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_0_groups_0 = const()[name = tensor<string, []>("conv_0_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [257, 1, 512]> expand_dims_1_to_fp16 = const()[name = tensor<string, []>("expand_dims_1_to_fp16"), val = tensor<fp16, [257, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1, 257, ?]> conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = tensor<string, []>("conv_0_cast_fp16")];
+            tensor<string, []> conv_1_pad_type_0 = const()[name = tensor<string, []>("conv_1_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_1_pad_0 = const()[name = tensor<string, []>("conv_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_1_dilations_0 = const()[name = tensor<string, []>("conv_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_1_groups_0 = const()[name = tensor<string, []>("conv_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [257, 1, 512]> expand_dims_2_to_fp16 = const()[name = tensor<string, []>("expand_dims_2_to_fp16"), val = tensor<fp16, [257, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263296)))];
+            tensor<fp16, [1, 257, ?]> conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = tensor<string, []>("conv_1_cast_fp16")];
+            tensor<int32, []> stack_0_axis_0 = const()[name = tensor<string, []>("stack_0_axis_0"), val = tensor<int32, []>(-1)];
+            tensor<fp16, [1, 257, ?, 2]> stack_0_cast_fp16 = stack(axis = stack_0_axis_0, values = (conv_0_cast_fp16, conv_1_cast_fp16))[name = tensor<string, []>("stack_0_cast_fp16")];
+            tensor<fp16, []> var_19_promoted_to_fp16 = const()[name = tensor<string, []>("op_19_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
+            tensor<fp16, [1, 257, ?, 2]> var_74_cast_fp16 = pow(x = stack_0_cast_fp16, y = var_19_promoted_to_fp16)[name = tensor<string, []>("op_74_cast_fp16")];
+            tensor<int32, [1]> var_76_axes_0 = const()[name = tensor<string, []>("op_76_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, []> var_76_keep_dims_0 = const()[name = tensor<string, []>("op_76_keep_dims_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 257, ?]> var_76_cast_fp16 = reduce_sum(axes = var_76_axes_0, keep_dims = var_76_keep_dims_0, x = var_74_cast_fp16)[name = tensor<string, []>("op_76_cast_fp16")];
+            tensor<fp16, [1, 257, ?]> x_11_cast_fp16 = identity(x = var_76_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")];
+            tensor<bool, []> x_13_transpose_x_0 = const()[name = tensor<string, []>("x_13_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_13_transpose_y_0 = const()[name = tensor<string, []>("x_13_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 128, 257]> const_4_to_fp16 = const()[name = tensor<string, []>("const_4_to_fp16"), val = tensor<fp16, [1, 128, 257]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526528)))];
+            tensor<fp16, [1, 128, ?]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = const_4_to_fp16, y = x_11_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
+            tensor<fp16, []> var_83_to_fp16 = const()[name = tensor<string, []>("op_83_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
+            tensor<fp16, [1, 128, ?]> var_84_cast_fp16 = add(x = x_13_cast_fp16, y = var_83_to_fp16)[name = tensor<string, []>("op_84_cast_fp16")];
+            tensor<fp32, []> x_epsilon_0 = const()[name = tensor<string, []>("x_epsilon_0"), val = tensor<fp32, []>(0x1p-149)];
+            tensor<fp16, [1, 128, ?]> x_cast_fp16 = log(epsilon = x_epsilon_0, x = var_84_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
+            tensor<int32, [3]> var_86_shape_cast_fp16 = shape(x = x_cast_fp16)[name = tensor<string, []>("op_86_shape_cast_fp16")];
+            tensor<int32, []> gather_5_axis_0 = const()[name = tensor<string, []>("gather_5_axis_0"), val = tensor<int32, []>(0)];
+            tensor<int32, []> gather_5_batch_dims_0 = const()[name = tensor<string, []>("gather_5_batch_dims_0"), val = tensor<int32, []>(0)];
+            tensor<bool, []> gather_5_validate_indices_0 = const()[name = tensor<string, []>("gather_5_validate_indices_0"), val = tensor<bool, []>(false)];
+            tensor<string, []> var_86_shape_cast_fp16_to_uint16_dtype_0 = const()[name = tensor<string, []>("op_86_shape_cast_fp16_to_uint16_dtype_0"), val = tensor<string, []>("uint16")];
+            tensor<uint16, []> select_5_to_uint16 = const()[name = tensor<string, []>("select_5_to_uint16"), val = tensor<uint16, []>(2)];
+            tensor<uint16, [3]> var_86_shape_cast_fp16_to_uint16 = cast(dtype = var_86_shape_cast_fp16_to_uint16_dtype_0, x = var_86_shape_cast_fp16)[name = tensor<string, []>("cast_11")];
+            tensor<uint16, []> gather_5_cast_uint16 = gather(axis = gather_5_axis_0, batch_dims = gather_5_batch_dims_0, indices = select_5_to_uint16, validate_indices = gather_5_validate_indices_0, x = var_86_shape_cast_fp16_to_uint16)[name = tensor<string, []>("gather_5_cast_uint16")];
+            tensor<string, []> gather_5_cast_uint16_to_int32_dtype_0 = const()[name = tensor<string, []>("gather_5_cast_uint16_to_int32_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, []> const_5 = const()[name = tensor<string, []>("const_5"), val = tensor<int32, []>(0)];
+            tensor<int32, []> const_6 = const()[name = tensor<string, []>("const_6"), val = tensor<int32, []>(1)];
+            tensor<int32, []> gather_5_cast_uint16_to_int32 = cast(dtype = gather_5_cast_uint16_to_int32_dtype_0, x = gather_5_cast_uint16)[name = tensor<string, []>("cast_10")];
+            tensor<int32, [?]> mask_1 = range_1d(end = gather_5_cast_uint16_to_int32, start = const_5, step = const_6)[name = tensor<string, []>("mask_1")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1, ?]> expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = mask_1)[name = tensor<string, []>("expand_dims_0")];
+            tensor<int32, [1]> var_91_axes_0 = const()[name = tensor<string, []>("op_91_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1, 1]> var_91 = expand_dims(axes = var_91_axes_0, x = mel_length)[name = tensor<string, []>("op_91")];
+            tensor<bool, [1, ?]> mask = greater_equal(x = expand_dims_0, y = var_91)[name = tensor<string, []>("mask")];
+            tensor<int32, [1]> var_93_axes_0 = const()[name = tensor<string, []>("op_93_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, ?]> var_93 = expand_dims(axes = var_93_axes_0, x = mask)[name = tensor<string, []>("op_93")];
+            tensor<fp16, [1, 128, ?]> processed_signal_cast_fp16 = select(a = var_16_to_fp16, b = x_cast_fp16, cond = var_93)[name = tensor<string, []>("processed_signal_cast_fp16")];
+            tensor<string, []> processed_signal_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("processed_signal_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 128, ?]> mel = cast(dtype = processed_signal_cast_fp16_to_fp32_dtype_0, x = processed_signal_cast_fp16)[name = tensor<string, []>("cast_9")];
+        } -> (mel, mel_length);
+}

ios17/multilingual/560ms/preprocessor.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:297514e2b211d14b0e53cb97193d679bb89ead98d28e578f3f1d049ddbcc36b3
+size 592384