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", "9.0"}})]
{
    func main<ios17>(tensor<fp32, [1, 1, 1001, 64]> input_features) {
            tensor<int32, []> var_18 = const()[name = tensor<string, []>("op_18"), val = tensor<int32, []>(-1)];
            tensor<int32, [4]> input_1_perm_0 = const()[name = tensor<string, []>("input_1_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<string, []> input_features_to_fp16_dtype_0 = const()[name = tensor<string, []>("input_features_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [64]> audio_model_audio_encoder_batch_norm_running_mean_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_batch_norm_running_mean_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [64]> audio_model_audio_encoder_batch_norm_running_var_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_batch_norm_running_var_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256)))];
            tensor<fp16, [64]> audio_model_audio_encoder_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_batch_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448)))];
            tensor<fp16, [64]> audio_model_audio_encoder_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_batch_norm_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640)))];
            tensor<fp16, []> var_7_to_fp16 = const()[name = tensor<string, []>("op_7_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1, 1001, 64]> input_features_to_fp16 = cast(dtype = input_features_to_fp16_dtype_0, x = input_features)[name = tensor<string, []>("cast_232")];
            tensor<fp16, [1, 64, 1001, 1]> input_1_cast_fp16 = transpose(perm = input_1_perm_0, x = input_features_to_fp16)[name = tensor<string, []>("transpose_165")];
            tensor<fp16, [1, 64, 1001, 1]> normalized_input_features_cast_fp16 = batch_norm(beta = audio_model_audio_encoder_batch_norm_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_batch_norm_weight_to_fp16, mean = audio_model_audio_encoder_batch_norm_running_mean_to_fp16, variance = audio_model_audio_encoder_batch_norm_running_var_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("normalized_input_features_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = tensor<string, []>("transpose_2_perm_0"), val = tensor<int32, [4]>([2, 0, 3, 1])];
            tensor<int32, [2]> concat_4 = const()[name = tensor<string, []>("concat_4"), val = tensor<int32, [2]>([1001, 64])];
            tensor<fp16, [1001, 1, 1, 64]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = normalized_input_features_cast_fp16)[name = tensor<string, []>("transpose_164")];
            tensor<fp16, [1001, 64]> reshape_1_cast_fp16 = reshape(shape = concat_4, x = transpose_2_cast_fp16)[name = tensor<string, []>("reshape_1_cast_fp16")];
            tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1024, 1001]> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, [1024, 1001]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(832)))];
            tensor<fp16, [1024, 64]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = const_0_to_fp16, y = reshape_1_cast_fp16)[name = tensor<string, []>("matmul_0_cast_fp16")];
            tensor<int32, [4]> concat_7 = const()[name = tensor<string, []>("concat_7"), val = tensor<int32, [4]>([1024, 1, 1, 64])];
            tensor<fp16, [1024, 1, 1, 64]> reshape_2_cast_fp16 = reshape(shape = concat_7, x = matmul_0_cast_fp16)[name = tensor<string, []>("reshape_2_cast_fp16")];
            tensor<int32, [4]> x_3_perm_0 = const()[name = tensor<string, []>("x_3_perm_0"), val = tensor<int32, [4]>([1, 2, 0, 3])];
            tensor<int32, [4]> var_77 = const()[name = tensor<string, []>("op_77"), val = tensor<int32, [4]>([1, 4, 256, 64])];
            tensor<fp16, [1, 1, 1024, 64]> x_3_cast_fp16 = transpose(perm = x_3_perm_0, x = reshape_2_cast_fp16)[name = tensor<string, []>("transpose_163")];
            tensor<fp16, [1, 4, 256, 64]> x_5_cast_fp16 = reshape(shape = var_77, x = x_3_cast_fp16)[name = tensor<string, []>("x_5_cast_fp16")];
            tensor<int32, [4]> var_79 = const()[name = tensor<string, []>("op_79"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_87 = const()[name = tensor<string, []>("op_87"), val = tensor<int32, [4]>([1, 1, 256, 256])];
            tensor<fp16, [1, 4, 64, 256]> var_80_cast_fp16 = transpose(perm = var_79, x = x_5_cast_fp16)[name = tensor<string, []>("transpose_162")];
            tensor<fp16, [1, 1, 256, 256]> hidden_states_1_cast_fp16 = reshape(shape = var_87, x = var_80_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")];
            tensor<string, []> hidden_states_3_pad_type_0 = const()[name = tensor<string, []>("hidden_states_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = tensor<string, []>("hidden_states_3_strides_0"), val = tensor<int32, [2]>([4, 4])];
            tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = tensor<string, []>("hidden_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = tensor<string, []>("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_3_groups_0 = const()[name = tensor<string, []>("hidden_states_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [96, 1, 4, 4]> audio_model_audio_encoder_patch_embed_proj_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_patch_embed_proj_weight_to_fp16"), val = tensor<fp16, [96, 1, 4, 4]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2050944)))];
            tensor<fp16, [96]> audio_model_audio_encoder_patch_embed_proj_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_patch_embed_proj_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2054080)))];
            tensor<fp16, [1, 96, 64, 64]> hidden_states_3_cast_fp16 = conv(bias = audio_model_audio_encoder_patch_embed_proj_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = audio_model_audio_encoder_patch_embed_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("hidden_states_3_cast_fp16")];
            tensor<int32, [3]> concat_8 = const()[name = tensor<string, []>("concat_8"), val = tensor<int32, [3]>([1, 96, 4096])];
            tensor<fp16, [1, 96, 4096]> var_100_cast_fp16 = reshape(shape = concat_8, x = hidden_states_3_cast_fp16)[name = tensor<string, []>("op_100_cast_fp16")];
            tensor<int32, [3]> input_3_perm_0 = const()[name = tensor<string, []>("input_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> hidden_states_5_axes_0 = const()[name = tensor<string, []>("hidden_states_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [96]> audio_model_audio_encoder_patch_embed_norm_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_patch_embed_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2054336)))];
            tensor<fp16, [96]> audio_model_audio_encoder_patch_embed_norm_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_patch_embed_norm_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2054592)))];
            tensor<fp16, [1, 4096, 96]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = var_100_cast_fp16)[name = tensor<string, []>("transpose_161")];
            tensor<fp16, [1, 4096, 96]> hidden_states_5_cast_fp16 = layer_norm(axes = hidden_states_5_axes_0, beta = audio_model_audio_encoder_patch_embed_norm_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_patch_embed_norm_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")];
            tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_layernorm_before_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2054848)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_layernorm_before_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2055104)))];
            tensor<fp16, [1, 4096, 96]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = audio_model_audio_encoder_layers_0_blocks_0_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_0_blocks_0_layernorm_before_weight_to_fp16, x = hidden_states_5_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")];
            tensor<int32, [4]> reshape_3_shape_0 = const()[name = tensor<string, []>("reshape_3_shape_0"), val = tensor<int32, [4]>([8, 8, 8, 768])];
            tensor<fp16, [8, 8, 8, 768]> reshape_3_cast_fp16 = reshape(shape = reshape_3_shape_0, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("reshape_3_cast_fp16")];
            tensor<int32, [4]> transpose_15_perm_0 = const()[name = tensor<string, []>("transpose_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_4_shape_0 = const()[name = tensor<string, []>("reshape_4_shape_0"), val = tensor<int32, [3]>([-1, 64, 96])];
            tensor<fp16, [8, 8, 8, 768]> transpose_15_cast_fp16 = transpose(perm = transpose_15_perm_0, x = reshape_3_cast_fp16)[name = tensor<string, []>("transpose_160")];
            tensor<fp16, [64, 64, 96]> reshape_4_cast_fp16 = reshape(shape = reshape_4_shape_0, x = transpose_15_cast_fp16)[name = tensor<string, []>("reshape_4_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_0_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_self_query_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2055360)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_self_query_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2073856)))];
            tensor<fp16, [64, 64, 96]> linear_0_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_0_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_0_attention_self_query_weight_to_fp16, x = reshape_4_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<int32, [4]> var_161 = const()[name = tensor<string, []>("op_161"), val = tensor<int32, [4]>([64, 64, -1, 24])];
            tensor<fp16, [64, 64, 4, 24]> var_162_cast_fp16 = reshape(shape = var_161, x = linear_0_cast_fp16)[name = tensor<string, []>("op_162_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_0_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_self_key_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2074112)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_self_key_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2092608)))];
            tensor<fp16, [64, 64, 96]> linear_1_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_0_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_0_attention_self_key_weight_to_fp16, x = reshape_4_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<int32, [4]> var_167 = const()[name = tensor<string, []>("op_167"), val = tensor<int32, [4]>([64, 64, -1, 24])];
            tensor<fp16, [64, 64, 4, 24]> var_168_cast_fp16 = reshape(shape = var_167, x = linear_1_cast_fp16)[name = tensor<string, []>("op_168_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_0_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_self_value_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2092864)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_self_value_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2111360)))];
            tensor<fp16, [64, 64, 96]> linear_2_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_0_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_0_attention_self_value_weight_to_fp16, x = reshape_4_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_173 = const()[name = tensor<string, []>("op_173"), val = tensor<int32, [4]>([64, 64, -1, 24])];
            tensor<fp16, [64, 64, 4, 24]> var_174_cast_fp16 = reshape(shape = var_173, x = linear_2_cast_fp16)[name = tensor<string, []>("op_174_cast_fp16")];
            tensor<int32, [4]> value_layer_1_perm_0 = const()[name = tensor<string, []>("value_layer_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_1_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_1_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_63_perm_0 = const()[name = tensor<string, []>("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_64_perm_0 = const()[name = tensor<string, []>("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [64, 4, 24, 64]> transpose_64 = transpose(perm = transpose_64_perm_0, x = var_168_cast_fp16)[name = tensor<string, []>("transpose_157")];
            tensor<fp16, [64, 4, 64, 24]> transpose_63 = transpose(perm = transpose_63_perm_0, x = var_162_cast_fp16)[name = tensor<string, []>("transpose_158")];
            tensor<fp16, [64, 4, 64, 64]> attention_scores_1_cast_fp16 = matmul(transpose_x = attention_scores_1_transpose_x_0, transpose_y = attention_scores_1_transpose_y_0, x = transpose_63, y = transpose_64)[name = tensor<string, []>("attention_scores_1_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_3_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_3_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [64, 4, 64, 64]> _inversed_attention_scores_3_cast_fp16 = mul(x = attention_scores_1_cast_fp16, y = _inversed_attention_scores_3_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_3_cast_fp16")];
            tensor<fp16, [1, 4, 64, 64]> var_189_to_fp16 = const()[name = tensor<string, []>("op_189_to_fp16"), val = tensor<fp16, [1, 4, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2111616)))];
            tensor<fp16, [64, 4, 64, 64]> input_7_cast_fp16 = add(x = _inversed_attention_scores_3_cast_fp16, y = var_189_to_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<fp16, [64, 4, 64, 64]> input_9_cast_fp16 = softmax(axis = var_18, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<bool, []> context_layer_1_transpose_x_0 = const()[name = tensor<string, []>("context_layer_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_1_transpose_y_0 = const()[name = tensor<string, []>("context_layer_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [64, 4, 64, 24]> value_layer_1_cast_fp16 = transpose(perm = value_layer_1_perm_0, x = var_174_cast_fp16)[name = tensor<string, []>("transpose_159")];
            tensor<fp16, [64, 4, 64, 24]> context_layer_1_cast_fp16 = matmul(transpose_x = context_layer_1_transpose_x_0, transpose_y = context_layer_1_transpose_y_0, x = input_9_cast_fp16, y = value_layer_1_cast_fp16)[name = tensor<string, []>("context_layer_1_cast_fp16")];
            tensor<int32, [4]> var_194 = const()[name = tensor<string, []>("op_194"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_199 = const()[name = tensor<string, []>("op_199"), val = tensor<int32, [3]>([64, 64, 96])];
            tensor<fp16, [64, 64, 4, 24]> var_195_cast_fp16 = transpose(perm = var_194, x = context_layer_1_cast_fp16)[name = tensor<string, []>("transpose_156")];
            tensor<fp16, [64, 64, 96]> input_11_cast_fp16 = reshape(shape = var_199, x = var_195_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_0_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2144448)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2162944)))];
            tensor<fp16, [64, 64, 96]> linear_3_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_0_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_0_attention_output_dense_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<int32, [4]> reshape_5_shape_0 = const()[name = tensor<string, []>("reshape_5_shape_0"), val = tensor<int32, [4]>([8, 8, 8, 768])];
            tensor<fp16, [8, 8, 8, 768]> reshape_5_cast_fp16 = reshape(shape = reshape_5_shape_0, x = linear_3_cast_fp16)[name = tensor<string, []>("reshape_5_cast_fp16")];
            tensor<int32, [4]> transpose_16_perm_0 = const()[name = tensor<string, []>("transpose_16_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_6_shape_0 = const()[name = tensor<string, []>("reshape_6_shape_0"), val = tensor<int32, [3]>([1, 4096, 96])];
            tensor<fp16, [8, 8, 8, 768]> transpose_16_cast_fp16 = transpose(perm = transpose_16_perm_0, x = reshape_5_cast_fp16)[name = tensor<string, []>("transpose_155")];
            tensor<fp16, [1, 4096, 96]> reshape_6_cast_fp16 = reshape(shape = reshape_6_shape_0, x = transpose_16_cast_fp16)[name = tensor<string, []>("reshape_6_cast_fp16")];
            tensor<fp16, [1, 4096, 96]> input_15_cast_fp16 = add(x = hidden_states_5_cast_fp16, y = reshape_6_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
            tensor<int32, [1]> input_17_axes_0 = const()[name = tensor<string, []>("input_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_layernorm_after_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2163200)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_layernorm_after_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2163456)))];
            tensor<fp16, [1, 4096, 96]> input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = audio_model_audio_encoder_layers_0_blocks_0_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_0_blocks_0_layernorm_after_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<fp16, [384, 96]> audio_model_audio_encoder_layers_0_blocks_0_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2163712)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_0_blocks_0_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2237504)))];
            tensor<fp16, [1, 4096, 384]> linear_4_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_0_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_0_intermediate_dense_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> input_21_mode_0 = const()[name = tensor<string, []>("input_21_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 384]> input_21_cast_fp16 = gelu(mode = input_21_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<fp16, [96, 384]> audio_model_audio_encoder_layers_0_blocks_0_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_output_dense_weight_to_fp16"), val = tensor<fp16, [96, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2238336)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_0_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_0_output_dense_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312128)))];
            tensor<fp16, [1, 4096, 96]> linear_5_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_0_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_0_output_dense_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 4096, 96]> hidden_states_15_cast_fp16 = add(x = input_15_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("hidden_states_15_cast_fp16")];
            tensor<int32, [1]> hidden_states_17_axes_0 = const()[name = tensor<string, []>("hidden_states_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_layernorm_before_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312384)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_layernorm_before_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312640)))];
            tensor<fp16, [1, 4096, 96]> hidden_states_17_cast_fp16 = layer_norm(axes = hidden_states_17_axes_0, beta = audio_model_audio_encoder_layers_0_blocks_1_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_0_blocks_1_layernorm_before_weight_to_fp16, x = hidden_states_15_cast_fp16)[name = tensor<string, []>("hidden_states_17_cast_fp16")];
            tensor<int32, [4]> var_249 = const()[name = tensor<string, []>("op_249"), val = tensor<int32, [4]>([1, 64, 64, 96])];
            tensor<fp16, [1, 64, 64, 96]> input_25_cast_fp16 = reshape(shape = var_249, x = hidden_states_17_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(1)];
            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<uint16, [64]> concat_9_to_uint16 = const()[name = tensor<string, []>("concat_9_to_uint16"), val = tensor<uint16, [64]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 0, 1, 2, 3])];
            tensor<fp16, [1, 64, 64, 96]> gather_0_cast_fp16_cast_uint16 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = concat_9_to_uint16, validate_indices = gather_0_validate_indices_0, x = input_25_cast_fp16)[name = tensor<string, []>("gather_0_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_1_axis_0 = const()[name = tensor<string, []>("gather_1_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_1_batch_dims_0 = const()[name = tensor<string, []>("gather_1_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_1_validate_indices_0 = const()[name = tensor<string, []>("gather_1_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [64]> concat_10_to_uint16 = const()[name = tensor<string, []>("concat_10_to_uint16"), val = tensor<uint16, [64]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 0, 1, 2, 3])];
            tensor<fp16, [1, 64, 64, 96]> gather_1_cast_fp16_cast_uint16 = gather(axis = gather_1_axis_0, batch_dims = gather_1_batch_dims_0, indices = concat_10_to_uint16, validate_indices = gather_1_validate_indices_0, x = gather_0_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_1_cast_fp16_cast_uint16")];
            tensor<int32, [4]> reshape_7_shape_0 = const()[name = tensor<string, []>("reshape_7_shape_0"), val = tensor<int32, [4]>([8, 8, 8, 768])];
            tensor<fp16, [8, 8, 8, 768]> reshape_7_cast_fp16 = reshape(shape = reshape_7_shape_0, x = gather_1_cast_fp16_cast_uint16)[name = tensor<string, []>("reshape_7_cast_fp16")];
            tensor<int32, [4]> transpose_17_perm_0 = const()[name = tensor<string, []>("transpose_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_8_shape_0 = const()[name = tensor<string, []>("reshape_8_shape_0"), val = tensor<int32, [3]>([-1, 64, 96])];
            tensor<fp16, [8, 8, 8, 768]> transpose_17_cast_fp16 = transpose(perm = transpose_17_perm_0, x = reshape_7_cast_fp16)[name = tensor<string, []>("transpose_154")];
            tensor<fp16, [64, 64, 96]> reshape_8_cast_fp16 = reshape(shape = reshape_8_shape_0, x = transpose_17_cast_fp16)[name = tensor<string, []>("reshape_8_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_1_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_self_query_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312896)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_self_query_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2331392)))];
            tensor<fp16, [64, 64, 96]> linear_6_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_1_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_1_attention_self_query_weight_to_fp16, x = reshape_8_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<int32, [4]> var_350 = const()[name = tensor<string, []>("op_350"), val = tensor<int32, [4]>([64, 64, -1, 24])];
            tensor<fp16, [64, 64, 4, 24]> var_351_cast_fp16 = reshape(shape = var_350, x = linear_6_cast_fp16)[name = tensor<string, []>("op_351_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_1_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_self_key_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2331648)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_self_key_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2350144)))];
            tensor<fp16, [64, 64, 96]> linear_7_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_1_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_1_attention_self_key_weight_to_fp16, x = reshape_8_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<int32, [4]> var_356 = const()[name = tensor<string, []>("op_356"), val = tensor<int32, [4]>([64, 64, -1, 24])];
            tensor<fp16, [64, 64, 4, 24]> var_357_cast_fp16 = reshape(shape = var_356, x = linear_7_cast_fp16)[name = tensor<string, []>("op_357_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_1_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_self_value_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2350400)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_self_value_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2368896)))];
            tensor<fp16, [64, 64, 96]> linear_8_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_1_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_1_attention_self_value_weight_to_fp16, x = reshape_8_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<int32, [4]> var_362 = const()[name = tensor<string, []>("op_362"), val = tensor<int32, [4]>([64, 64, -1, 24])];
            tensor<fp16, [64, 64, 4, 24]> var_363_cast_fp16 = reshape(shape = var_362, x = linear_8_cast_fp16)[name = tensor<string, []>("op_363_cast_fp16")];
            tensor<int32, [4]> value_layer_3_perm_0 = const()[name = tensor<string, []>("value_layer_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_5_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_5_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_65_perm_0 = const()[name = tensor<string, []>("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_66_perm_0 = const()[name = tensor<string, []>("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [64, 4, 24, 64]> transpose_66 = transpose(perm = transpose_66_perm_0, x = var_357_cast_fp16)[name = tensor<string, []>("transpose_151")];
            tensor<fp16, [64, 4, 64, 24]> transpose_65 = transpose(perm = transpose_65_perm_0, x = var_351_cast_fp16)[name = tensor<string, []>("transpose_152")];
            tensor<fp16, [64, 4, 64, 64]> attention_scores_5_cast_fp16 = matmul(transpose_x = attention_scores_5_transpose_x_0, transpose_y = attention_scores_5_transpose_y_0, x = transpose_65, y = transpose_66)[name = tensor<string, []>("attention_scores_5_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_7_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_7_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [64, 4, 64, 64]> _inversed_attention_scores_7_cast_fp16 = mul(x = attention_scores_5_cast_fp16, y = _inversed_attention_scores_7_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_7_cast_fp16")];
            tensor<fp16, [1, 4, 64, 64]> var_378_to_fp16 = const()[name = tensor<string, []>("op_378_to_fp16"), val = tensor<fp16, [1, 4, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2369152)))];
            tensor<fp16, [64, 4, 64, 64]> attention_scores_9_cast_fp16 = add(x = _inversed_attention_scores_7_cast_fp16, y = var_378_to_fp16)[name = tensor<string, []>("attention_scores_9_cast_fp16")];
            tensor<int32, [5]> var_384 = const()[name = tensor<string, []>("op_384"), val = tensor<int32, [5]>([1, 64, 4, 64, 64])];
            tensor<fp16, [1, 64, 4, 64, 64]> attention_scores_11_cast_fp16 = reshape(shape = var_384, x = attention_scores_9_cast_fp16)[name = tensor<string, []>("attention_scores_11_cast_fp16")];
            tensor<fp16, [1, 64, 1, 64, 64]> var_387_to_fp16 = const()[name = tensor<string, []>("op_387_to_fp16"), val = tensor<fp16, [1, 64, 1, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2401984)))];
            tensor<fp16, [1, 64, 4, 64, 64]> attention_scores_13_cast_fp16 = add(x = attention_scores_11_cast_fp16, y = var_387_to_fp16)[name = tensor<string, []>("attention_scores_13_cast_fp16")];
            tensor<int32, [4]> var_389 = const()[name = tensor<string, []>("op_389"), val = tensor<int32, [4]>([-1, 4, 64, 64])];
            tensor<fp16, [64, 4, 64, 64]> input_27_cast_fp16 = reshape(shape = var_389, x = attention_scores_13_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
            tensor<fp16, [64, 4, 64, 64]> input_29_cast_fp16 = softmax(axis = var_18, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
            tensor<bool, []> context_layer_5_transpose_x_0 = const()[name = tensor<string, []>("context_layer_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_5_transpose_y_0 = const()[name = tensor<string, []>("context_layer_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [64, 4, 64, 24]> value_layer_3_cast_fp16 = transpose(perm = value_layer_3_perm_0, x = var_363_cast_fp16)[name = tensor<string, []>("transpose_153")];
            tensor<fp16, [64, 4, 64, 24]> context_layer_5_cast_fp16 = matmul(transpose_x = context_layer_5_transpose_x_0, transpose_y = context_layer_5_transpose_y_0, x = input_29_cast_fp16, y = value_layer_3_cast_fp16)[name = tensor<string, []>("context_layer_5_cast_fp16")];
            tensor<int32, [4]> var_394 = const()[name = tensor<string, []>("op_394"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_399 = const()[name = tensor<string, []>("op_399"), val = tensor<int32, [3]>([64, 64, 96])];
            tensor<fp16, [64, 64, 4, 24]> var_395_cast_fp16 = transpose(perm = var_394, x = context_layer_5_cast_fp16)[name = tensor<string, []>("transpose_150")];
            tensor<fp16, [64, 64, 96]> input_31_cast_fp16 = reshape(shape = var_399, x = var_395_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<fp16, [96, 96]> audio_model_audio_encoder_layers_0_blocks_1_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [96, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2926336)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2944832)))];
            tensor<fp16, [64, 64, 96]> linear_9_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_1_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_1_attention_output_dense_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<int32, [4]> reshape_9_shape_0 = const()[name = tensor<string, []>("reshape_9_shape_0"), val = tensor<int32, [4]>([8, 8, 8, 768])];
            tensor<fp16, [8, 8, 8, 768]> reshape_9_cast_fp16 = reshape(shape = reshape_9_shape_0, x = linear_9_cast_fp16)[name = tensor<string, []>("reshape_9_cast_fp16")];
            tensor<int32, [4]> transpose_18_perm_0 = const()[name = tensor<string, []>("transpose_18_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> reshape_10_shape_0 = const()[name = tensor<string, []>("reshape_10_shape_0"), val = tensor<int32, [4]>([-1, 64, 64, 96])];
            tensor<fp16, [8, 8, 8, 768]> transpose_18_cast_fp16 = transpose(perm = transpose_18_perm_0, x = reshape_9_cast_fp16)[name = tensor<string, []>("transpose_149")];
            tensor<fp16, [1, 64, 64, 96]> reshape_10_cast_fp16 = reshape(shape = reshape_10_shape_0, x = transpose_18_cast_fp16)[name = tensor<string, []>("reshape_10_cast_fp16")];
            tensor<int32, []> gather_2_axis_0 = const()[name = tensor<string, []>("gather_2_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_2_batch_dims_0 = const()[name = tensor<string, []>("gather_2_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_2_validate_indices_0 = const()[name = tensor<string, []>("gather_2_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [64]> concat_11_to_uint16 = const()[name = tensor<string, []>("concat_11_to_uint16"), val = tensor<uint16, [64]>([60, 61, 62, 63, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59])];
            tensor<fp16, [1, 64, 64, 96]> gather_2_cast_fp16_cast_uint16 = gather(axis = gather_2_axis_0, batch_dims = gather_2_batch_dims_0, indices = concat_11_to_uint16, validate_indices = gather_2_validate_indices_0, x = reshape_10_cast_fp16)[name = tensor<string, []>("gather_2_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_3_axis_0 = const()[name = tensor<string, []>("gather_3_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_3_batch_dims_0 = const()[name = tensor<string, []>("gather_3_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_3_validate_indices_0 = const()[name = tensor<string, []>("gather_3_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [64]> concat_12_to_uint16 = const()[name = tensor<string, []>("concat_12_to_uint16"), val = tensor<uint16, [64]>([60, 61, 62, 63, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59])];
            tensor<fp16, [1, 64, 64, 96]> gather_3_cast_fp16_cast_uint16 = gather(axis = gather_3_axis_0, batch_dims = gather_3_batch_dims_0, indices = concat_12_to_uint16, validate_indices = gather_3_validate_indices_0, x = gather_2_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_3_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_423 = const()[name = tensor<string, []>("op_423"), val = tensor<int32, [3]>([1, 4096, 96])];
            tensor<fp16, [1, 4096, 96]> attention_windows_7_cast_fp16 = reshape(shape = var_423, x = gather_3_cast_fp16_cast_uint16)[name = tensor<string, []>("attention_windows_7_cast_fp16")];
            tensor<fp16, [1, 4096, 96]> input_35_cast_fp16 = add(x = hidden_states_15_cast_fp16, y = attention_windows_7_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<int32, [1]> input_37_axes_0 = const()[name = tensor<string, []>("input_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_layernorm_after_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2945088)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_layernorm_after_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2945344)))];
            tensor<fp16, [1, 4096, 96]> input_37_cast_fp16 = layer_norm(axes = input_37_axes_0, beta = audio_model_audio_encoder_layers_0_blocks_1_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_0_blocks_1_layernorm_after_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<fp16, [384, 96]> audio_model_audio_encoder_layers_0_blocks_1_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [384, 96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2945600)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_0_blocks_1_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3019392)))];
            tensor<fp16, [1, 4096, 384]> linear_10_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_1_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_1_intermediate_dense_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<string, []> input_41_mode_0 = const()[name = tensor<string, []>("input_41_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 384]> input_41_cast_fp16 = gelu(mode = input_41_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<fp16, [96, 384]> audio_model_audio_encoder_layers_0_blocks_1_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_output_dense_weight_to_fp16"), val = tensor<fp16, [96, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3020224)))];
            tensor<fp16, [96]> audio_model_audio_encoder_layers_0_blocks_1_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_blocks_1_output_dense_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3094016)))];
            tensor<fp16, [1, 4096, 96]> linear_11_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_0_blocks_1_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_0_blocks_1_output_dense_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, [1, 4096, 96]> input_feature_1_cast_fp16 = add(x = input_35_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_feature_1_cast_fp16")];
            tensor<int32, [4]> var_446 = const()[name = tensor<string, []>("op_446"), val = tensor<int32, [4]>([1, 64, 64, 96])];
            tensor<fp16, [1, 64, 64, 96]> input_feature_3_cast_fp16 = reshape(shape = var_446, x = input_feature_1_cast_fp16)[name = tensor<string, []>("input_feature_3_cast_fp16")];
            tensor<int32, [4]> var_449_begin_0 = const()[name = tensor<string, []>("op_449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_449_end_0 = const()[name = tensor<string, []>("op_449_end_0"), val = tensor<int32, [4]>([1, 64, 64, 96])];
            tensor<bool, [4]> var_449_end_mask_0 = const()[name = tensor<string, []>("op_449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_449_stride_0 = const()[name = tensor<string, []>("op_449_stride_0"), val = tensor<int32, [4]>([1, 2, 1, 1])];
            tensor<fp16, [1, 32, 64, 96]> var_449_cast_fp16 = slice_by_index(begin = var_449_begin_0, end = var_449_end_0, end_mask = var_449_end_mask_0, stride = var_449_stride_0, x = input_feature_3_cast_fp16)[name = tensor<string, []>("op_449_cast_fp16")];
            tensor<int32, [4]> var_450_begin_0 = const()[name = tensor<string, []>("op_450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_450_end_0 = const()[name = tensor<string, []>("op_450_end_0"), val = tensor<int32, [4]>([1, 32, 64, 96])];
            tensor<bool, [4]> var_450_end_mask_0 = const()[name = tensor<string, []>("op_450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_450_stride_0 = const()[name = tensor<string, []>("op_450_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 32, 32, 96]> var_450_cast_fp16 = slice_by_index(begin = var_450_begin_0, end = var_450_end_0, end_mask = var_450_end_mask_0, stride = var_450_stride_0, x = var_449_cast_fp16)[name = tensor<string, []>("op_450_cast_fp16")];
            tensor<int32, [4]> var_453_begin_0 = const()[name = tensor<string, []>("op_453_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])];
            tensor<int32, [4]> var_453_end_0 = const()[name = tensor<string, []>("op_453_end_0"), val = tensor<int32, [4]>([1, 64, 64, 96])];
            tensor<bool, [4]> var_453_end_mask_0 = const()[name = tensor<string, []>("op_453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_453_stride_0 = const()[name = tensor<string, []>("op_453_stride_0"), val = tensor<int32, [4]>([1, 2, 1, 1])];
            tensor<fp16, [1, 32, 64, 96]> var_453_cast_fp16 = slice_by_index(begin = var_453_begin_0, end = var_453_end_0, end_mask = var_453_end_mask_0, stride = var_453_stride_0, x = input_feature_3_cast_fp16)[name = tensor<string, []>("op_453_cast_fp16")];
            tensor<int32, [4]> var_454_begin_0 = const()[name = tensor<string, []>("op_454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_454_end_0 = const()[name = tensor<string, []>("op_454_end_0"), val = tensor<int32, [4]>([1, 32, 64, 96])];
            tensor<bool, [4]> var_454_end_mask_0 = const()[name = tensor<string, []>("op_454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_454_stride_0 = const()[name = tensor<string, []>("op_454_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 32, 32, 96]> var_454_cast_fp16 = slice_by_index(begin = var_454_begin_0, end = var_454_end_0, end_mask = var_454_end_mask_0, stride = var_454_stride_0, x = var_453_cast_fp16)[name = tensor<string, []>("op_454_cast_fp16")];
            tensor<int32, [4]> var_458_begin_0 = const()[name = tensor<string, []>("op_458_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_458_end_0 = const()[name = tensor<string, []>("op_458_end_0"), val = tensor<int32, [4]>([1, 32, 64, 96])];
            tensor<bool, [4]> var_458_end_mask_0 = const()[name = tensor<string, []>("op_458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_458_stride_0 = const()[name = tensor<string, []>("op_458_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 32, 32, 96]> var_458_cast_fp16 = slice_by_index(begin = var_458_begin_0, end = var_458_end_0, end_mask = var_458_end_mask_0, stride = var_458_stride_0, x = var_449_cast_fp16)[name = tensor<string, []>("op_458_cast_fp16")];
            tensor<int32, [4]> var_462_begin_0 = const()[name = tensor<string, []>("op_462_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_462_end_0 = const()[name = tensor<string, []>("op_462_end_0"), val = tensor<int32, [4]>([1, 32, 64, 96])];
            tensor<bool, [4]> var_462_end_mask_0 = const()[name = tensor<string, []>("op_462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_462_stride_0 = const()[name = tensor<string, []>("op_462_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 32, 32, 96]> var_462_cast_fp16 = slice_by_index(begin = var_462_begin_0, end = var_462_end_0, end_mask = var_462_end_mask_0, stride = var_462_stride_0, x = var_453_cast_fp16)[name = tensor<string, []>("op_462_cast_fp16")];
            tensor<bool, []> input_feature_5_interleave_0 = const()[name = tensor<string, []>("input_feature_5_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 32, 384]> input_feature_5_cast_fp16 = concat(axis = var_18, interleave = input_feature_5_interleave_0, values = (var_450_cast_fp16, var_454_cast_fp16, var_458_cast_fp16, var_462_cast_fp16))[name = tensor<string, []>("input_feature_5_cast_fp16")];
            tensor<int32, [3]> var_469 = const()[name = tensor<string, []>("op_469"), val = tensor<int32, [3]>([1, -1, 384])];
            tensor<fp16, [1, 1024, 384]> input_45_cast_fp16 = reshape(shape = var_469, x = input_feature_5_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<int32, [1]> input_47_axes_0 = const()[name = tensor<string, []>("input_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_0_downsample_norm_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_downsample_norm_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3094272)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_0_downsample_norm_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_downsample_norm_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3095104)))];
            tensor<fp16, [1, 1024, 384]> input_47_cast_fp16 = layer_norm(axes = input_47_axes_0, beta = audio_model_audio_encoder_layers_0_downsample_norm_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_0_downsample_norm_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<fp16, [192, 384]> audio_model_audio_encoder_layers_0_downsample_reduction_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_0_downsample_reduction_weight_to_fp16"), val = tensor<fp16, [192, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3095936)))];
            tensor<fp16, [192]> linear_12_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_12_bias_0_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3243456)))];
            tensor<fp16, [1, 1024, 192]> linear_12_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = audio_model_audio_encoder_layers_0_downsample_reduction_weight_to_fp16, x = input_47_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<int32, [1]> hidden_states_33_axes_0 = const()[name = tensor<string, []>("hidden_states_33_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_layernorm_before_weight_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3243904)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_layernorm_before_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3244352)))];
            tensor<fp16, [1, 1024, 192]> hidden_states_33_cast_fp16 = layer_norm(axes = hidden_states_33_axes_0, beta = audio_model_audio_encoder_layers_1_blocks_0_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_1_blocks_0_layernorm_before_weight_to_fp16, x = linear_12_cast_fp16)[name = tensor<string, []>("hidden_states_33_cast_fp16")];
            tensor<int32, [4]> reshape_11_shape_0 = const()[name = tensor<string, []>("reshape_11_shape_0"), val = tensor<int32, [4]>([4, 8, 4, 1536])];
            tensor<fp16, [4, 8, 4, 1536]> reshape_11_cast_fp16 = reshape(shape = reshape_11_shape_0, x = hidden_states_33_cast_fp16)[name = tensor<string, []>("reshape_11_cast_fp16")];
            tensor<int32, [4]> transpose_19_perm_0 = const()[name = tensor<string, []>("transpose_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_12_shape_0 = const()[name = tensor<string, []>("reshape_12_shape_0"), val = tensor<int32, [3]>([-1, 64, 192])];
            tensor<fp16, [4, 4, 8, 1536]> transpose_19_cast_fp16 = transpose(perm = transpose_19_perm_0, x = reshape_11_cast_fp16)[name = tensor<string, []>("transpose_148")];
            tensor<fp16, [16, 64, 192]> reshape_12_cast_fp16 = reshape(shape = reshape_12_shape_0, x = transpose_19_cast_fp16)[name = tensor<string, []>("reshape_12_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_0_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_self_query_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3244800)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_self_query_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3318592)))];
            tensor<fp16, [16, 64, 192]> linear_13_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_0_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_0_attention_self_query_weight_to_fp16, x = reshape_12_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<int32, [4]> var_532 = const()[name = tensor<string, []>("op_532"), val = tensor<int32, [4]>([16, 64, -1, 24])];
            tensor<fp16, [16, 64, 8, 24]> var_533_cast_fp16 = reshape(shape = var_532, x = linear_13_cast_fp16)[name = tensor<string, []>("op_533_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_0_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_self_key_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3319040)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_self_key_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3392832)))];
            tensor<fp16, [16, 64, 192]> linear_14_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_0_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_0_attention_self_key_weight_to_fp16, x = reshape_12_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_538 = const()[name = tensor<string, []>("op_538"), val = tensor<int32, [4]>([16, 64, -1, 24])];
            tensor<fp16, [16, 64, 8, 24]> var_539_cast_fp16 = reshape(shape = var_538, x = linear_14_cast_fp16)[name = tensor<string, []>("op_539_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_0_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_self_value_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3393280)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_self_value_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3467072)))];
            tensor<fp16, [16, 64, 192]> linear_15_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_0_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_0_attention_self_value_weight_to_fp16, x = reshape_12_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<int32, [4]> var_544 = const()[name = tensor<string, []>("op_544"), val = tensor<int32, [4]>([16, 64, -1, 24])];
            tensor<fp16, [16, 64, 8, 24]> var_545_cast_fp16 = reshape(shape = var_544, x = linear_15_cast_fp16)[name = tensor<string, []>("op_545_cast_fp16")];
            tensor<int32, [4]> value_layer_5_perm_0 = const()[name = tensor<string, []>("value_layer_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_15_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_15_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_15_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_15_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_67_perm_0 = const()[name = tensor<string, []>("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_68_perm_0 = const()[name = tensor<string, []>("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [16, 8, 24, 64]> transpose_68 = transpose(perm = transpose_68_perm_0, x = var_539_cast_fp16)[name = tensor<string, []>("transpose_145")];
            tensor<fp16, [16, 8, 64, 24]> transpose_67 = transpose(perm = transpose_67_perm_0, x = var_533_cast_fp16)[name = tensor<string, []>("transpose_146")];
            tensor<fp16, [16, 8, 64, 64]> attention_scores_15_cast_fp16 = matmul(transpose_x = attention_scores_15_transpose_x_0, transpose_y = attention_scores_15_transpose_y_0, x = transpose_67, y = transpose_68)[name = tensor<string, []>("attention_scores_15_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_17_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [16, 8, 64, 64]> _inversed_attention_scores_17_cast_fp16 = mul(x = attention_scores_15_cast_fp16, y = _inversed_attention_scores_17_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_17_cast_fp16")];
            tensor<fp16, [1, 8, 64, 64]> var_560_to_fp16 = const()[name = tensor<string, []>("op_560_to_fp16"), val = tensor<fp16, [1, 8, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3467520)))];
            tensor<fp16, [16, 8, 64, 64]> input_51_cast_fp16 = add(x = _inversed_attention_scores_17_cast_fp16, y = var_560_to_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
            tensor<fp16, [16, 8, 64, 64]> input_53_cast_fp16 = softmax(axis = var_18, x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<bool, []> context_layer_9_transpose_x_0 = const()[name = tensor<string, []>("context_layer_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_9_transpose_y_0 = const()[name = tensor<string, []>("context_layer_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [16, 8, 64, 24]> value_layer_5_cast_fp16 = transpose(perm = value_layer_5_perm_0, x = var_545_cast_fp16)[name = tensor<string, []>("transpose_147")];
            tensor<fp16, [16, 8, 64, 24]> context_layer_9_cast_fp16 = matmul(transpose_x = context_layer_9_transpose_x_0, transpose_y = context_layer_9_transpose_y_0, x = input_53_cast_fp16, y = value_layer_5_cast_fp16)[name = tensor<string, []>("context_layer_9_cast_fp16")];
            tensor<int32, [4]> var_565 = const()[name = tensor<string, []>("op_565"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_570 = const()[name = tensor<string, []>("op_570"), val = tensor<int32, [3]>([16, 64, 192])];
            tensor<fp16, [16, 64, 8, 24]> var_566_cast_fp16 = transpose(perm = var_565, x = context_layer_9_cast_fp16)[name = tensor<string, []>("transpose_144")];
            tensor<fp16, [16, 64, 192]> input_55_cast_fp16 = reshape(shape = var_570, x = var_566_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_0_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3533120)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3606912)))];
            tensor<fp16, [16, 64, 192]> linear_16_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_0_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_0_attention_output_dense_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<int32, [4]> reshape_13_shape_0 = const()[name = tensor<string, []>("reshape_13_shape_0"), val = tensor<int32, [4]>([4, 4, 8, 1536])];
            tensor<fp16, [4, 4, 8, 1536]> reshape_13_cast_fp16 = reshape(shape = reshape_13_shape_0, x = linear_16_cast_fp16)[name = tensor<string, []>("reshape_13_cast_fp16")];
            tensor<int32, [4]> transpose_20_perm_0 = const()[name = tensor<string, []>("transpose_20_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_14_shape_0 = const()[name = tensor<string, []>("reshape_14_shape_0"), val = tensor<int32, [3]>([1, 1024, 192])];
            tensor<fp16, [4, 8, 4, 1536]> transpose_20_cast_fp16 = transpose(perm = transpose_20_perm_0, x = reshape_13_cast_fp16)[name = tensor<string, []>("transpose_143")];
            tensor<fp16, [1, 1024, 192]> reshape_14_cast_fp16 = reshape(shape = reshape_14_shape_0, x = transpose_20_cast_fp16)[name = tensor<string, []>("reshape_14_cast_fp16")];
            tensor<fp16, [1, 1024, 192]> input_59_cast_fp16 = add(x = linear_12_cast_fp16, y = reshape_14_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_layernorm_after_weight_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3607360)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_layernorm_after_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3607808)))];
            tensor<fp16, [1, 1024, 192]> input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = audio_model_audio_encoder_layers_1_blocks_0_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_1_blocks_0_layernorm_after_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<fp16, [768, 192]> audio_model_audio_encoder_layers_1_blocks_0_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [768, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3608256)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_1_blocks_0_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3903232)))];
            tensor<fp16, [1, 1024, 768]> linear_17_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_0_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_0_intermediate_dense_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<string, []> input_65_mode_0 = const()[name = tensor<string, []>("input_65_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 768]> input_65_cast_fp16 = gelu(mode = input_65_mode_0, x = linear_17_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
            tensor<fp16, [192, 768]> audio_model_audio_encoder_layers_1_blocks_0_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_output_dense_weight_to_fp16"), val = tensor<fp16, [192, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3904832)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_0_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_0_output_dense_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4199808)))];
            tensor<fp16, [1, 1024, 192]> linear_18_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_0_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_0_output_dense_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<fp16, [1, 1024, 192]> hidden_states_41_cast_fp16 = add(x = input_59_cast_fp16, y = linear_18_cast_fp16)[name = tensor<string, []>("hidden_states_41_cast_fp16")];
            tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_layernorm_before_weight_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4200256)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_layernorm_before_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4200704)))];
            tensor<fp16, [1, 1024, 192]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = audio_model_audio_encoder_layers_1_blocks_1_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_1_blocks_1_layernorm_before_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")];
            tensor<int32, [4]> var_620 = const()[name = tensor<string, []>("op_620"), val = tensor<int32, [4]>([1, 32, 32, 192])];
            tensor<fp16, [1, 32, 32, 192]> input_69_cast_fp16 = reshape(shape = var_620, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
            tensor<int32, []> gather_4_axis_0 = const()[name = tensor<string, []>("gather_4_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_4_batch_dims_0 = const()[name = tensor<string, []>("gather_4_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_4_validate_indices_0 = const()[name = tensor<string, []>("gather_4_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [32]> concat_13_to_uint16 = const()[name = tensor<string, []>("concat_13_to_uint16"), val = tensor<uint16, [32]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 0, 1, 2, 3])];
            tensor<fp16, [1, 32, 32, 192]> gather_4_cast_fp16_cast_uint16 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = concat_13_to_uint16, validate_indices = gather_4_validate_indices_0, x = input_69_cast_fp16)[name = tensor<string, []>("gather_4_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_5_axis_0 = const()[name = tensor<string, []>("gather_5_axis_0"), val = tensor<int32, []>(2)];
            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<uint16, [32]> concat_14_to_uint16 = const()[name = tensor<string, []>("concat_14_to_uint16"), val = tensor<uint16, [32]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 0, 1, 2, 3])];
            tensor<fp16, [1, 32, 32, 192]> gather_5_cast_fp16_cast_uint16 = gather(axis = gather_5_axis_0, batch_dims = gather_5_batch_dims_0, indices = concat_14_to_uint16, validate_indices = gather_5_validate_indices_0, x = gather_4_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_5_cast_fp16_cast_uint16")];
            tensor<int32, [4]> reshape_15_shape_0 = const()[name = tensor<string, []>("reshape_15_shape_0"), val = tensor<int32, [4]>([4, 8, 4, 1536])];
            tensor<fp16, [4, 8, 4, 1536]> reshape_15_cast_fp16 = reshape(shape = reshape_15_shape_0, x = gather_5_cast_fp16_cast_uint16)[name = tensor<string, []>("reshape_15_cast_fp16")];
            tensor<int32, [4]> transpose_21_perm_0 = const()[name = tensor<string, []>("transpose_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_16_shape_0 = const()[name = tensor<string, []>("reshape_16_shape_0"), val = tensor<int32, [3]>([-1, 64, 192])];
            tensor<fp16, [4, 4, 8, 1536]> transpose_21_cast_fp16 = transpose(perm = transpose_21_perm_0, x = reshape_15_cast_fp16)[name = tensor<string, []>("transpose_142")];
            tensor<fp16, [16, 64, 192]> reshape_16_cast_fp16 = reshape(shape = reshape_16_shape_0, x = transpose_21_cast_fp16)[name = tensor<string, []>("reshape_16_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_1_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_self_query_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4201152)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_self_query_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4274944)))];
            tensor<fp16, [16, 64, 192]> linear_19_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_1_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_1_attention_self_query_weight_to_fp16, x = reshape_16_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<int32, [4]> var_721 = const()[name = tensor<string, []>("op_721"), val = tensor<int32, [4]>([16, 64, -1, 24])];
            tensor<fp16, [16, 64, 8, 24]> var_722_cast_fp16 = reshape(shape = var_721, x = linear_19_cast_fp16)[name = tensor<string, []>("op_722_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_1_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_self_key_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4275392)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_self_key_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4349184)))];
            tensor<fp16, [16, 64, 192]> linear_20_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_1_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_1_attention_self_key_weight_to_fp16, x = reshape_16_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_727 = const()[name = tensor<string, []>("op_727"), val = tensor<int32, [4]>([16, 64, -1, 24])];
            tensor<fp16, [16, 64, 8, 24]> var_728_cast_fp16 = reshape(shape = var_727, x = linear_20_cast_fp16)[name = tensor<string, []>("op_728_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_1_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_self_value_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4349632)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_self_value_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4423424)))];
            tensor<fp16, [16, 64, 192]> linear_21_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_1_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_1_attention_self_value_weight_to_fp16, x = reshape_16_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<int32, [4]> var_733 = const()[name = tensor<string, []>("op_733"), val = tensor<int32, [4]>([16, 64, -1, 24])];
            tensor<fp16, [16, 64, 8, 24]> var_734_cast_fp16 = reshape(shape = var_733, x = linear_21_cast_fp16)[name = tensor<string, []>("op_734_cast_fp16")];
            tensor<int32, [4]> value_layer_7_perm_0 = const()[name = tensor<string, []>("value_layer_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_19_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_19_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_19_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_19_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_69_perm_0 = const()[name = tensor<string, []>("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_70_perm_0 = const()[name = tensor<string, []>("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [16, 8, 24, 64]> transpose_70 = transpose(perm = transpose_70_perm_0, x = var_728_cast_fp16)[name = tensor<string, []>("transpose_139")];
            tensor<fp16, [16, 8, 64, 24]> transpose_69 = transpose(perm = transpose_69_perm_0, x = var_722_cast_fp16)[name = tensor<string, []>("transpose_140")];
            tensor<fp16, [16, 8, 64, 64]> attention_scores_19_cast_fp16 = matmul(transpose_x = attention_scores_19_transpose_x_0, transpose_y = attention_scores_19_transpose_y_0, x = transpose_69, y = transpose_70)[name = tensor<string, []>("attention_scores_19_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_21_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_21_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [16, 8, 64, 64]> _inversed_attention_scores_21_cast_fp16 = mul(x = attention_scores_19_cast_fp16, y = _inversed_attention_scores_21_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_21_cast_fp16")];
            tensor<fp16, [1, 8, 64, 64]> var_749_to_fp16 = const()[name = tensor<string, []>("op_749_to_fp16"), val = tensor<fp16, [1, 8, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4423872)))];
            tensor<fp16, [16, 8, 64, 64]> attention_scores_23_cast_fp16 = add(x = _inversed_attention_scores_21_cast_fp16, y = var_749_to_fp16)[name = tensor<string, []>("attention_scores_23_cast_fp16")];
            tensor<int32, [5]> var_755 = const()[name = tensor<string, []>("op_755"), val = tensor<int32, [5]>([1, 16, 8, 64, 64])];
            tensor<fp16, [1, 16, 8, 64, 64]> attention_scores_25_cast_fp16 = reshape(shape = var_755, x = attention_scores_23_cast_fp16)[name = tensor<string, []>("attention_scores_25_cast_fp16")];
            tensor<fp16, [1, 16, 1, 64, 64]> var_758_to_fp16 = const()[name = tensor<string, []>("op_758_to_fp16"), val = tensor<fp16, [1, 16, 1, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4489472)))];
            tensor<fp16, [1, 16, 8, 64, 64]> attention_scores_27_cast_fp16 = add(x = attention_scores_25_cast_fp16, y = var_758_to_fp16)[name = tensor<string, []>("attention_scores_27_cast_fp16")];
            tensor<int32, [4]> var_760 = const()[name = tensor<string, []>("op_760"), val = tensor<int32, [4]>([-1, 8, 64, 64])];
            tensor<fp16, [16, 8, 64, 64]> input_71_cast_fp16 = reshape(shape = var_760, x = attention_scores_27_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
            tensor<fp16, [16, 8, 64, 64]> input_73_cast_fp16 = softmax(axis = var_18, x = input_71_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<bool, []> context_layer_13_transpose_x_0 = const()[name = tensor<string, []>("context_layer_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_13_transpose_y_0 = const()[name = tensor<string, []>("context_layer_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [16, 8, 64, 24]> value_layer_7_cast_fp16 = transpose(perm = value_layer_7_perm_0, x = var_734_cast_fp16)[name = tensor<string, []>("transpose_141")];
            tensor<fp16, [16, 8, 64, 24]> context_layer_13_cast_fp16 = matmul(transpose_x = context_layer_13_transpose_x_0, transpose_y = context_layer_13_transpose_y_0, x = input_73_cast_fp16, y = value_layer_7_cast_fp16)[name = tensor<string, []>("context_layer_13_cast_fp16")];
            tensor<int32, [4]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_770 = const()[name = tensor<string, []>("op_770"), val = tensor<int32, [3]>([16, 64, 192])];
            tensor<fp16, [16, 64, 8, 24]> var_766_cast_fp16 = transpose(perm = var_765, x = context_layer_13_cast_fp16)[name = tensor<string, []>("transpose_138")];
            tensor<fp16, [16, 64, 192]> input_75_cast_fp16 = reshape(shape = var_770, x = var_766_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
            tensor<fp16, [192, 192]> audio_model_audio_encoder_layers_1_blocks_1_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [192, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4620608)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4694400)))];
            tensor<fp16, [16, 64, 192]> linear_22_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_1_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_1_attention_output_dense_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<int32, [4]> reshape_17_shape_0 = const()[name = tensor<string, []>("reshape_17_shape_0"), val = tensor<int32, [4]>([4, 4, 8, 1536])];
            tensor<fp16, [4, 4, 8, 1536]> reshape_17_cast_fp16 = reshape(shape = reshape_17_shape_0, x = linear_22_cast_fp16)[name = tensor<string, []>("reshape_17_cast_fp16")];
            tensor<int32, [4]> transpose_22_perm_0 = const()[name = tensor<string, []>("transpose_22_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> reshape_18_shape_0 = const()[name = tensor<string, []>("reshape_18_shape_0"), val = tensor<int32, [4]>([-1, 32, 32, 192])];
            tensor<fp16, [4, 8, 4, 1536]> transpose_22_cast_fp16 = transpose(perm = transpose_22_perm_0, x = reshape_17_cast_fp16)[name = tensor<string, []>("transpose_137")];
            tensor<fp16, [1, 32, 32, 192]> reshape_18_cast_fp16 = reshape(shape = reshape_18_shape_0, x = transpose_22_cast_fp16)[name = tensor<string, []>("reshape_18_cast_fp16")];
            tensor<int32, []> gather_6_axis_0 = const()[name = tensor<string, []>("gather_6_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_6_batch_dims_0 = const()[name = tensor<string, []>("gather_6_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_6_validate_indices_0 = const()[name = tensor<string, []>("gather_6_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [32]> concat_15_to_uint16 = const()[name = tensor<string, []>("concat_15_to_uint16"), val = tensor<uint16, [32]>([28, 29, 30, 31, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27])];
            tensor<fp16, [1, 32, 32, 192]> gather_6_cast_fp16_cast_uint16 = gather(axis = gather_6_axis_0, batch_dims = gather_6_batch_dims_0, indices = concat_15_to_uint16, validate_indices = gather_6_validate_indices_0, x = reshape_18_cast_fp16)[name = tensor<string, []>("gather_6_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_7_axis_0 = const()[name = tensor<string, []>("gather_7_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_7_batch_dims_0 = const()[name = tensor<string, []>("gather_7_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_7_validate_indices_0 = const()[name = tensor<string, []>("gather_7_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [32]> concat_16_to_uint16 = const()[name = tensor<string, []>("concat_16_to_uint16"), val = tensor<uint16, [32]>([28, 29, 30, 31, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27])];
            tensor<fp16, [1, 32, 32, 192]> gather_7_cast_fp16_cast_uint16 = gather(axis = gather_7_axis_0, batch_dims = gather_7_batch_dims_0, indices = concat_16_to_uint16, validate_indices = gather_7_validate_indices_0, x = gather_6_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_7_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_794 = const()[name = tensor<string, []>("op_794"), val = tensor<int32, [3]>([1, 1024, 192])];
            tensor<fp16, [1, 1024, 192]> attention_windows_15_cast_fp16 = reshape(shape = var_794, x = gather_7_cast_fp16_cast_uint16)[name = tensor<string, []>("attention_windows_15_cast_fp16")];
            tensor<fp16, [1, 1024, 192]> input_79_cast_fp16 = add(x = hidden_states_41_cast_fp16, y = attention_windows_15_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")];
            tensor<int32, [1]> input_81_axes_0 = const()[name = tensor<string, []>("input_81_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_layernorm_after_weight_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4694848)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_layernorm_after_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4695296)))];
            tensor<fp16, [1, 1024, 192]> input_81_cast_fp16 = layer_norm(axes = input_81_axes_0, beta = audio_model_audio_encoder_layers_1_blocks_1_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_1_blocks_1_layernorm_after_weight_to_fp16, x = input_79_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")];
            tensor<fp16, [768, 192]> audio_model_audio_encoder_layers_1_blocks_1_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [768, 192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4695744)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_1_blocks_1_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4990720)))];
            tensor<fp16, [1, 1024, 768]> linear_23_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_1_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_1_intermediate_dense_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<string, []> input_85_mode_0 = const()[name = tensor<string, []>("input_85_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 768]> input_85_cast_fp16 = gelu(mode = input_85_mode_0, x = linear_23_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<fp16, [192, 768]> audio_model_audio_encoder_layers_1_blocks_1_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_output_dense_weight_to_fp16"), val = tensor<fp16, [192, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4992320)))];
            tensor<fp16, [192]> audio_model_audio_encoder_layers_1_blocks_1_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_blocks_1_output_dense_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5287296)))];
            tensor<fp16, [1, 1024, 192]> linear_24_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_1_blocks_1_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_1_blocks_1_output_dense_weight_to_fp16, x = input_85_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<fp16, [1, 1024, 192]> input_feature_7_cast_fp16 = add(x = input_79_cast_fp16, y = linear_24_cast_fp16)[name = tensor<string, []>("input_feature_7_cast_fp16")];
            tensor<int32, [4]> var_817 = const()[name = tensor<string, []>("op_817"), val = tensor<int32, [4]>([1, 32, 32, 192])];
            tensor<fp16, [1, 32, 32, 192]> input_feature_9_cast_fp16 = reshape(shape = var_817, x = input_feature_7_cast_fp16)[name = tensor<string, []>("input_feature_9_cast_fp16")];
            tensor<int32, [4]> var_820_begin_0 = const()[name = tensor<string, []>("op_820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_820_end_0 = const()[name = tensor<string, []>("op_820_end_0"), val = tensor<int32, [4]>([1, 32, 32, 192])];
            tensor<bool, [4]> var_820_end_mask_0 = const()[name = tensor<string, []>("op_820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_820_stride_0 = const()[name = tensor<string, []>("op_820_stride_0"), val = tensor<int32, [4]>([1, 2, 1, 1])];
            tensor<fp16, [1, 16, 32, 192]> var_820_cast_fp16 = slice_by_index(begin = var_820_begin_0, end = var_820_end_0, end_mask = var_820_end_mask_0, stride = var_820_stride_0, x = input_feature_9_cast_fp16)[name = tensor<string, []>("op_820_cast_fp16")];
            tensor<int32, [4]> var_821_begin_0 = const()[name = tensor<string, []>("op_821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_821_end_0 = const()[name = tensor<string, []>("op_821_end_0"), val = tensor<int32, [4]>([1, 16, 32, 192])];
            tensor<bool, [4]> var_821_end_mask_0 = const()[name = tensor<string, []>("op_821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_821_stride_0 = const()[name = tensor<string, []>("op_821_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 16, 16, 192]> var_821_cast_fp16 = slice_by_index(begin = var_821_begin_0, end = var_821_end_0, end_mask = var_821_end_mask_0, stride = var_821_stride_0, x = var_820_cast_fp16)[name = tensor<string, []>("op_821_cast_fp16")];
            tensor<int32, [4]> var_824_begin_0 = const()[name = tensor<string, []>("op_824_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])];
            tensor<int32, [4]> var_824_end_0 = const()[name = tensor<string, []>("op_824_end_0"), val = tensor<int32, [4]>([1, 32, 32, 192])];
            tensor<bool, [4]> var_824_end_mask_0 = const()[name = tensor<string, []>("op_824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_824_stride_0 = const()[name = tensor<string, []>("op_824_stride_0"), val = tensor<int32, [4]>([1, 2, 1, 1])];
            tensor<fp16, [1, 16, 32, 192]> var_824_cast_fp16 = slice_by_index(begin = var_824_begin_0, end = var_824_end_0, end_mask = var_824_end_mask_0, stride = var_824_stride_0, x = input_feature_9_cast_fp16)[name = tensor<string, []>("op_824_cast_fp16")];
            tensor<int32, [4]> var_825_begin_0 = const()[name = tensor<string, []>("op_825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_825_end_0 = const()[name = tensor<string, []>("op_825_end_0"), val = tensor<int32, [4]>([1, 16, 32, 192])];
            tensor<bool, [4]> var_825_end_mask_0 = const()[name = tensor<string, []>("op_825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_825_stride_0 = const()[name = tensor<string, []>("op_825_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 16, 16, 192]> var_825_cast_fp16 = slice_by_index(begin = var_825_begin_0, end = var_825_end_0, end_mask = var_825_end_mask_0, stride = var_825_stride_0, x = var_824_cast_fp16)[name = tensor<string, []>("op_825_cast_fp16")];
            tensor<int32, [4]> var_829_begin_0 = const()[name = tensor<string, []>("op_829_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_829_end_0 = const()[name = tensor<string, []>("op_829_end_0"), val = tensor<int32, [4]>([1, 16, 32, 192])];
            tensor<bool, [4]> var_829_end_mask_0 = const()[name = tensor<string, []>("op_829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_829_stride_0 = const()[name = tensor<string, []>("op_829_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 16, 16, 192]> var_829_cast_fp16 = slice_by_index(begin = var_829_begin_0, end = var_829_end_0, end_mask = var_829_end_mask_0, stride = var_829_stride_0, x = var_820_cast_fp16)[name = tensor<string, []>("op_829_cast_fp16")];
            tensor<int32, [4]> var_833_begin_0 = const()[name = tensor<string, []>("op_833_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_833_end_0 = const()[name = tensor<string, []>("op_833_end_0"), val = tensor<int32, [4]>([1, 16, 32, 192])];
            tensor<bool, [4]> var_833_end_mask_0 = const()[name = tensor<string, []>("op_833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_833_stride_0 = const()[name = tensor<string, []>("op_833_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 16, 16, 192]> var_833_cast_fp16 = slice_by_index(begin = var_833_begin_0, end = var_833_end_0, end_mask = var_833_end_mask_0, stride = var_833_stride_0, x = var_824_cast_fp16)[name = tensor<string, []>("op_833_cast_fp16")];
            tensor<bool, []> input_feature_11_interleave_0 = const()[name = tensor<string, []>("input_feature_11_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 16, 768]> input_feature_11_cast_fp16 = concat(axis = var_18, interleave = input_feature_11_interleave_0, values = (var_821_cast_fp16, var_825_cast_fp16, var_829_cast_fp16, var_833_cast_fp16))[name = tensor<string, []>("input_feature_11_cast_fp16")];
            tensor<int32, [3]> var_840 = const()[name = tensor<string, []>("op_840"), val = tensor<int32, [3]>([1, -1, 768])];
            tensor<fp16, [1, 256, 768]> input_89_cast_fp16 = reshape(shape = var_840, x = input_feature_11_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")];
            tensor<int32, [1]> input_91_axes_0 = const()[name = tensor<string, []>("input_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_1_downsample_norm_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_downsample_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5287744)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_1_downsample_norm_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_downsample_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5289344)))];
            tensor<fp16, [1, 256, 768]> input_91_cast_fp16 = layer_norm(axes = input_91_axes_0, beta = audio_model_audio_encoder_layers_1_downsample_norm_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_1_downsample_norm_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<fp16, [384, 768]> audio_model_audio_encoder_layers_1_downsample_reduction_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_1_downsample_reduction_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5290944)))];
            tensor<fp16, [384]> linear_25_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_25_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5880832)))];
            tensor<fp16, [1, 256, 384]> linear_25_cast_fp16 = linear(bias = linear_25_bias_0_to_fp16, weight = audio_model_audio_encoder_layers_1_downsample_reduction_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<int32, [1]> hidden_states_59_axes_0 = const()[name = tensor<string, []>("hidden_states_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_layernorm_before_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5881664)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_layernorm_before_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5882496)))];
            tensor<fp16, [1, 256, 384]> hidden_states_59_cast_fp16 = layer_norm(axes = hidden_states_59_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_0_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_0_layernorm_before_weight_to_fp16, x = linear_25_cast_fp16)[name = tensor<string, []>("hidden_states_59_cast_fp16")];
            tensor<int32, [4]> reshape_19_shape_0 = const()[name = tensor<string, []>("reshape_19_shape_0"), val = tensor<int32, [4]>([2, 8, 2, 3072])];
            tensor<fp16, [2, 8, 2, 3072]> reshape_19_cast_fp16 = reshape(shape = reshape_19_shape_0, x = hidden_states_59_cast_fp16)[name = tensor<string, []>("reshape_19_cast_fp16")];
            tensor<int32, [4]> transpose_23_perm_0 = const()[name = tensor<string, []>("transpose_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_20_shape_0 = const()[name = tensor<string, []>("reshape_20_shape_0"), val = tensor<int32, [3]>([-1, 64, 384])];
            tensor<fp16, [2, 2, 8, 3072]> transpose_23_cast_fp16 = transpose(perm = transpose_23_perm_0, x = reshape_19_cast_fp16)[name = tensor<string, []>("transpose_136")];
            tensor<fp16, [4, 64, 384]> reshape_20_cast_fp16 = reshape(shape = reshape_20_shape_0, x = transpose_23_cast_fp16)[name = tensor<string, []>("reshape_20_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_0_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_self_query_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5883328)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_self_query_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6178304)))];
            tensor<fp16, [4, 64, 384]> linear_26_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_0_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_0_attention_self_query_weight_to_fp16, x = reshape_20_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_911 = const()[name = tensor<string, []>("op_911"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_912_cast_fp16 = reshape(shape = var_911, x = linear_26_cast_fp16)[name = tensor<string, []>("op_912_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_0_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_self_key_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6179136)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_self_key_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6474112)))];
            tensor<fp16, [4, 64, 384]> linear_27_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_0_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_0_attention_self_key_weight_to_fp16, x = reshape_20_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<int32, [4]> var_917 = const()[name = tensor<string, []>("op_917"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_918_cast_fp16 = reshape(shape = var_917, x = linear_27_cast_fp16)[name = tensor<string, []>("op_918_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_0_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_self_value_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6474944)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_self_value_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6769920)))];
            tensor<fp16, [4, 64, 384]> linear_28_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_0_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_0_attention_self_value_weight_to_fp16, x = reshape_20_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<int32, [4]> var_923 = const()[name = tensor<string, []>("op_923"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_924_cast_fp16 = reshape(shape = var_923, x = linear_28_cast_fp16)[name = tensor<string, []>("op_924_cast_fp16")];
            tensor<int32, [4]> value_layer_9_perm_0 = const()[name = tensor<string, []>("value_layer_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_29_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_29_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_71_perm_0 = const()[name = tensor<string, []>("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [4, 16, 24, 64]> transpose_72 = transpose(perm = transpose_72_perm_0, x = var_918_cast_fp16)[name = tensor<string, []>("transpose_133")];
            tensor<fp16, [4, 16, 64, 24]> transpose_71 = transpose(perm = transpose_71_perm_0, x = var_912_cast_fp16)[name = tensor<string, []>("transpose_134")];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_29_cast_fp16 = matmul(transpose_x = attention_scores_29_transpose_x_0, transpose_y = attention_scores_29_transpose_y_0, x = transpose_71, y = transpose_72)[name = tensor<string, []>("attention_scores_29_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_31_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_31_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [4, 16, 64, 64]> _inversed_attention_scores_31_cast_fp16 = mul(x = attention_scores_29_cast_fp16, y = _inversed_attention_scores_31_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_31_cast_fp16")];
            tensor<fp16, [1, 16, 64, 64]> var_939_to_fp16 = const()[name = tensor<string, []>("op_939_to_fp16"), val = tensor<fp16, [1, 16, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6770752)))];
            tensor<fp16, [4, 16, 64, 64]> input_95_cast_fp16 = add(x = _inversed_attention_scores_31_cast_fp16, y = var_939_to_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<fp16, [4, 16, 64, 64]> input_97_cast_fp16 = softmax(axis = var_18, x = input_95_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<bool, []> context_layer_17_transpose_x_0 = const()[name = tensor<string, []>("context_layer_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_17_transpose_y_0 = const()[name = tensor<string, []>("context_layer_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [4, 16, 64, 24]> value_layer_9_cast_fp16 = transpose(perm = value_layer_9_perm_0, x = var_924_cast_fp16)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [4, 16, 64, 24]> context_layer_17_cast_fp16 = matmul(transpose_x = context_layer_17_transpose_x_0, transpose_y = context_layer_17_transpose_y_0, x = input_97_cast_fp16, y = value_layer_9_cast_fp16)[name = tensor<string, []>("context_layer_17_cast_fp16")];
            tensor<int32, [4]> var_944 = const()[name = tensor<string, []>("op_944"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_949 = const()[name = tensor<string, []>("op_949"), val = tensor<int32, [3]>([4, 64, 384])];
            tensor<fp16, [4, 64, 16, 24]> var_945_cast_fp16 = transpose(perm = var_944, x = context_layer_17_cast_fp16)[name = tensor<string, []>("transpose_132")];
            tensor<fp16, [4, 64, 384]> input_99_cast_fp16 = reshape(shape = var_949, x = var_945_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_0_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6901888)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7196864)))];
            tensor<fp16, [4, 64, 384]> linear_29_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_0_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_0_attention_output_dense_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<int32, [4]> reshape_21_shape_0 = const()[name = tensor<string, []>("reshape_21_shape_0"), val = tensor<int32, [4]>([2, 2, 8, 3072])];
            tensor<fp16, [2, 2, 8, 3072]> reshape_21_cast_fp16 = reshape(shape = reshape_21_shape_0, x = linear_29_cast_fp16)[name = tensor<string, []>("reshape_21_cast_fp16")];
            tensor<int32, [4]> transpose_24_perm_0 = const()[name = tensor<string, []>("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_22_shape_0 = const()[name = tensor<string, []>("reshape_22_shape_0"), val = tensor<int32, [3]>([1, 256, 384])];
            tensor<fp16, [2, 8, 2, 3072]> transpose_24_cast_fp16 = transpose(perm = transpose_24_perm_0, x = reshape_21_cast_fp16)[name = tensor<string, []>("transpose_131")];
            tensor<fp16, [1, 256, 384]> reshape_22_cast_fp16 = reshape(shape = reshape_22_shape_0, x = transpose_24_cast_fp16)[name = tensor<string, []>("reshape_22_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_103_cast_fp16 = add(x = linear_25_cast_fp16, y = reshape_22_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<int32, [1]> input_105_axes_0 = const()[name = tensor<string, []>("input_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_layernorm_after_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7197696)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_layernorm_after_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7198528)))];
            tensor<fp16, [1, 256, 384]> input_105_cast_fp16 = layer_norm(axes = input_105_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_0_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_0_layernorm_after_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<fp16, [1536, 384]> audio_model_audio_encoder_layers_2_blocks_0_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7199360)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_blocks_0_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8379072)))];
            tensor<fp16, [1, 256, 1536]> linear_30_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_0_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_0_intermediate_dense_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<string, []> input_109_mode_0 = const()[name = tensor<string, []>("input_109_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 256, 1536]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = linear_30_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<fp16, [384, 1536]> audio_model_audio_encoder_layers_2_blocks_0_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8382208)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_0_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_0_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9561920)))];
            tensor<fp16, [1, 256, 384]> linear_31_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_0_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_0_output_dense_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<fp16, [1, 256, 384]> hidden_states_67_cast_fp16 = add(x = input_103_cast_fp16, y = linear_31_cast_fp16)[name = tensor<string, []>("hidden_states_67_cast_fp16")];
            tensor<int32, [1]> hidden_states_69_axes_0 = const()[name = tensor<string, []>("hidden_states_69_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_layernorm_before_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9562752)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_layernorm_before_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9563584)))];
            tensor<fp16, [1, 256, 384]> hidden_states_69_cast_fp16 = layer_norm(axes = hidden_states_69_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_1_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_1_layernorm_before_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("hidden_states_69_cast_fp16")];
            tensor<int32, [4]> var_999 = const()[name = tensor<string, []>("op_999"), val = tensor<int32, [4]>([1, 16, 16, 384])];
            tensor<fp16, [1, 16, 16, 384]> input_113_cast_fp16 = reshape(shape = var_999, x = hidden_states_69_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")];
            tensor<int32, []> gather_8_axis_0 = const()[name = tensor<string, []>("gather_8_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_8_batch_dims_0 = const()[name = tensor<string, []>("gather_8_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_8_validate_indices_0 = const()[name = tensor<string, []>("gather_8_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_17_to_uint16 = const()[name = tensor<string, []>("concat_17_to_uint16"), val = tensor<uint16, [16]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3])];
            tensor<fp16, [1, 16, 16, 384]> gather_8_cast_fp16_cast_uint16 = gather(axis = gather_8_axis_0, batch_dims = gather_8_batch_dims_0, indices = concat_17_to_uint16, validate_indices = gather_8_validate_indices_0, x = input_113_cast_fp16)[name = tensor<string, []>("gather_8_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_9_axis_0 = const()[name = tensor<string, []>("gather_9_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_9_batch_dims_0 = const()[name = tensor<string, []>("gather_9_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_9_validate_indices_0 = const()[name = tensor<string, []>("gather_9_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_18_to_uint16 = const()[name = tensor<string, []>("concat_18_to_uint16"), val = tensor<uint16, [16]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3])];
            tensor<fp16, [1, 16, 16, 384]> gather_9_cast_fp16_cast_uint16 = gather(axis = gather_9_axis_0, batch_dims = gather_9_batch_dims_0, indices = concat_18_to_uint16, validate_indices = gather_9_validate_indices_0, x = gather_8_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_9_cast_fp16_cast_uint16")];
            tensor<int32, [4]> reshape_23_shape_0 = const()[name = tensor<string, []>("reshape_23_shape_0"), val = tensor<int32, [4]>([2, 8, 2, 3072])];
            tensor<fp16, [2, 8, 2, 3072]> reshape_23_cast_fp16 = reshape(shape = reshape_23_shape_0, x = gather_9_cast_fp16_cast_uint16)[name = tensor<string, []>("reshape_23_cast_fp16")];
            tensor<int32, [4]> transpose_25_perm_0 = const()[name = tensor<string, []>("transpose_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_24_shape_0 = const()[name = tensor<string, []>("reshape_24_shape_0"), val = tensor<int32, [3]>([-1, 64, 384])];
            tensor<fp16, [2, 2, 8, 3072]> transpose_25_cast_fp16 = transpose(perm = transpose_25_perm_0, x = reshape_23_cast_fp16)[name = tensor<string, []>("transpose_130")];
            tensor<fp16, [4, 64, 384]> reshape_24_cast_fp16 = reshape(shape = reshape_24_shape_0, x = transpose_25_cast_fp16)[name = tensor<string, []>("reshape_24_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_1_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_self_query_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9564416)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_self_query_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9859392)))];
            tensor<fp16, [4, 64, 384]> linear_32_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_1_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_1_attention_self_query_weight_to_fp16, x = reshape_24_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_1100 = const()[name = tensor<string, []>("op_1100"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1101_cast_fp16 = reshape(shape = var_1100, x = linear_32_cast_fp16)[name = tensor<string, []>("op_1101_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_1_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_self_key_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9860224)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_self_key_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10155200)))];
            tensor<fp16, [4, 64, 384]> linear_33_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_1_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_1_attention_self_key_weight_to_fp16, x = reshape_24_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<int32, [4]> var_1106 = const()[name = tensor<string, []>("op_1106"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1107_cast_fp16 = reshape(shape = var_1106, x = linear_33_cast_fp16)[name = tensor<string, []>("op_1107_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_1_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_self_value_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10156032)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_self_value_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10451008)))];
            tensor<fp16, [4, 64, 384]> linear_34_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_1_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_1_attention_self_value_weight_to_fp16, x = reshape_24_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<int32, [4]> var_1112 = const()[name = tensor<string, []>("op_1112"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1113_cast_fp16 = reshape(shape = var_1112, x = linear_34_cast_fp16)[name = tensor<string, []>("op_1113_cast_fp16")];
            tensor<int32, [4]> value_layer_11_perm_0 = const()[name = tensor<string, []>("value_layer_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_33_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_33_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [4, 16, 24, 64]> transpose_74 = transpose(perm = transpose_74_perm_0, x = var_1107_cast_fp16)[name = tensor<string, []>("transpose_127")];
            tensor<fp16, [4, 16, 64, 24]> transpose_73 = transpose(perm = transpose_73_perm_0, x = var_1101_cast_fp16)[name = tensor<string, []>("transpose_128")];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_33_cast_fp16 = matmul(transpose_x = attention_scores_33_transpose_x_0, transpose_y = attention_scores_33_transpose_y_0, x = transpose_73, y = transpose_74)[name = tensor<string, []>("attention_scores_33_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_35_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_35_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [4, 16, 64, 64]> _inversed_attention_scores_35_cast_fp16 = mul(x = attention_scores_33_cast_fp16, y = _inversed_attention_scores_35_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_35_cast_fp16")];
            tensor<fp16, [1, 16, 64, 64]> var_1128_to_fp16 = const()[name = tensor<string, []>("op_1128_to_fp16"), val = tensor<fp16, [1, 16, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10451840)))];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_37_cast_fp16 = add(x = _inversed_attention_scores_35_cast_fp16, y = var_1128_to_fp16)[name = tensor<string, []>("attention_scores_37_cast_fp16")];
            tensor<int32, [5]> var_1134 = const()[name = tensor<string, []>("op_1134"), val = tensor<int32, [5]>([1, 4, 16, 64, 64])];
            tensor<fp16, [1, 4, 16, 64, 64]> attention_scores_39_cast_fp16 = reshape(shape = var_1134, x = attention_scores_37_cast_fp16)[name = tensor<string, []>("attention_scores_39_cast_fp16")];
            tensor<fp16, [1, 4, 1, 64, 64]> var_1137_to_fp16 = const()[name = tensor<string, []>("op_1137_to_fp16"), val = tensor<fp16, [1, 4, 1, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10582976)))];
            tensor<fp16, [1, 4, 16, 64, 64]> attention_scores_41_cast_fp16 = add(x = attention_scores_39_cast_fp16, y = var_1137_to_fp16)[name = tensor<string, []>("attention_scores_41_cast_fp16")];
            tensor<int32, [4]> var_1139 = const()[name = tensor<string, []>("op_1139"), val = tensor<int32, [4]>([-1, 16, 64, 64])];
            tensor<fp16, [4, 16, 64, 64]> input_115_cast_fp16 = reshape(shape = var_1139, x = attention_scores_41_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<fp16, [4, 16, 64, 64]> input_117_cast_fp16 = softmax(axis = var_18, x = input_115_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")];
            tensor<bool, []> context_layer_21_transpose_x_0 = const()[name = tensor<string, []>("context_layer_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_21_transpose_y_0 = const()[name = tensor<string, []>("context_layer_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [4, 16, 64, 24]> value_layer_11_cast_fp16 = transpose(perm = value_layer_11_perm_0, x = var_1113_cast_fp16)[name = tensor<string, []>("transpose_129")];
            tensor<fp16, [4, 16, 64, 24]> context_layer_21_cast_fp16 = matmul(transpose_x = context_layer_21_transpose_x_0, transpose_y = context_layer_21_transpose_y_0, x = input_117_cast_fp16, y = value_layer_11_cast_fp16)[name = tensor<string, []>("context_layer_21_cast_fp16")];
            tensor<int32, [4]> var_1144 = const()[name = tensor<string, []>("op_1144"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1149 = const()[name = tensor<string, []>("op_1149"), val = tensor<int32, [3]>([4, 64, 384])];
            tensor<fp16, [4, 64, 16, 24]> var_1145_cast_fp16 = transpose(perm = var_1144, x = context_layer_21_cast_fp16)[name = tensor<string, []>("transpose_126")];
            tensor<fp16, [4, 64, 384]> input_119_cast_fp16 = reshape(shape = var_1149, x = var_1145_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_1_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10615808)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10910784)))];
            tensor<fp16, [4, 64, 384]> linear_35_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_1_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_1_attention_output_dense_weight_to_fp16, x = input_119_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<int32, [4]> reshape_25_shape_0 = const()[name = tensor<string, []>("reshape_25_shape_0"), val = tensor<int32, [4]>([2, 2, 8, 3072])];
            tensor<fp16, [2, 2, 8, 3072]> reshape_25_cast_fp16 = reshape(shape = reshape_25_shape_0, x = linear_35_cast_fp16)[name = tensor<string, []>("reshape_25_cast_fp16")];
            tensor<int32, [4]> transpose_26_perm_0 = const()[name = tensor<string, []>("transpose_26_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> reshape_26_shape_0 = const()[name = tensor<string, []>("reshape_26_shape_0"), val = tensor<int32, [4]>([-1, 16, 16, 384])];
            tensor<fp16, [2, 8, 2, 3072]> transpose_26_cast_fp16 = transpose(perm = transpose_26_perm_0, x = reshape_25_cast_fp16)[name = tensor<string, []>("transpose_125")];
            tensor<fp16, [1, 16, 16, 384]> reshape_26_cast_fp16 = reshape(shape = reshape_26_shape_0, x = transpose_26_cast_fp16)[name = tensor<string, []>("reshape_26_cast_fp16")];
            tensor<int32, []> gather_10_axis_0 = const()[name = tensor<string, []>("gather_10_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_10_batch_dims_0 = const()[name = tensor<string, []>("gather_10_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_10_validate_indices_0 = const()[name = tensor<string, []>("gather_10_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_19_to_uint16 = const()[name = tensor<string, []>("concat_19_to_uint16"), val = tensor<uint16, [16]>([12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])];
            tensor<fp16, [1, 16, 16, 384]> gather_10_cast_fp16_cast_uint16 = gather(axis = gather_10_axis_0, batch_dims = gather_10_batch_dims_0, indices = concat_19_to_uint16, validate_indices = gather_10_validate_indices_0, x = reshape_26_cast_fp16)[name = tensor<string, []>("gather_10_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_11_axis_0 = const()[name = tensor<string, []>("gather_11_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_11_batch_dims_0 = const()[name = tensor<string, []>("gather_11_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_11_validate_indices_0 = const()[name = tensor<string, []>("gather_11_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_20_to_uint16 = const()[name = tensor<string, []>("concat_20_to_uint16"), val = tensor<uint16, [16]>([12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])];
            tensor<fp16, [1, 16, 16, 384]> gather_11_cast_fp16_cast_uint16 = gather(axis = gather_11_axis_0, batch_dims = gather_11_batch_dims_0, indices = concat_20_to_uint16, validate_indices = gather_11_validate_indices_0, x = gather_10_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_11_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_1173 = const()[name = tensor<string, []>("op_1173"), val = tensor<int32, [3]>([1, 256, 384])];
            tensor<fp16, [1, 256, 384]> attention_windows_23_cast_fp16 = reshape(shape = var_1173, x = gather_11_cast_fp16_cast_uint16)[name = tensor<string, []>("attention_windows_23_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_123_cast_fp16 = add(x = hidden_states_67_cast_fp16, y = attention_windows_23_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
            tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_layernorm_after_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10911616)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_layernorm_after_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10912448)))];
            tensor<fp16, [1, 256, 384]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_1_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_1_layernorm_after_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<fp16, [1536, 384]> audio_model_audio_encoder_layers_2_blocks_1_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10913280)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_blocks_1_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12092992)))];
            tensor<fp16, [1, 256, 1536]> linear_36_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_1_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_1_intermediate_dense_weight_to_fp16, x = input_125_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<string, []> input_129_mode_0 = const()[name = tensor<string, []>("input_129_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 256, 1536]> input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = linear_36_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
            tensor<fp16, [384, 1536]> audio_model_audio_encoder_layers_2_blocks_1_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12096128)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_1_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_1_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13275840)))];
            tensor<fp16, [1, 256, 384]> linear_37_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_1_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_1_output_dense_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1, 256, 384]> hidden_states_83_cast_fp16 = add(x = input_123_cast_fp16, y = linear_37_cast_fp16)[name = tensor<string, []>("hidden_states_83_cast_fp16")];
            tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = tensor<string, []>("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_layernorm_before_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13276672)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_layernorm_before_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13277504)))];
            tensor<fp16, [1, 256, 384]> hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_2_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_2_layernorm_before_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor<string, []>("hidden_states_85_cast_fp16")];
            tensor<int32, [4]> reshape_27_shape_0 = const()[name = tensor<string, []>("reshape_27_shape_0"), val = tensor<int32, [4]>([2, 8, 2, 3072])];
            tensor<fp16, [2, 8, 2, 3072]> reshape_27_cast_fp16 = reshape(shape = reshape_27_shape_0, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("reshape_27_cast_fp16")];
            tensor<int32, [4]> transpose_27_perm_0 = const()[name = tensor<string, []>("transpose_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_28_shape_0 = const()[name = tensor<string, []>("reshape_28_shape_0"), val = tensor<int32, [3]>([-1, 64, 384])];
            tensor<fp16, [2, 2, 8, 3072]> transpose_27_cast_fp16 = transpose(perm = transpose_27_perm_0, x = reshape_27_cast_fp16)[name = tensor<string, []>("transpose_124")];
            tensor<fp16, [4, 64, 384]> reshape_28_cast_fp16 = reshape(shape = reshape_28_shape_0, x = transpose_27_cast_fp16)[name = tensor<string, []>("reshape_28_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_2_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_self_query_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13278336)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_self_query_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13573312)))];
            tensor<fp16, [4, 64, 384]> linear_38_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_2_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_2_attention_self_query_weight_to_fp16, x = reshape_28_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_1241 = const()[name = tensor<string, []>("op_1241"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1242_cast_fp16 = reshape(shape = var_1241, x = linear_38_cast_fp16)[name = tensor<string, []>("op_1242_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_2_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_self_key_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13574144)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_self_key_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13869120)))];
            tensor<fp16, [4, 64, 384]> linear_39_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_2_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_2_attention_self_key_weight_to_fp16, x = reshape_28_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<int32, [4]> var_1247 = const()[name = tensor<string, []>("op_1247"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1248_cast_fp16 = reshape(shape = var_1247, x = linear_39_cast_fp16)[name = tensor<string, []>("op_1248_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_2_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_self_value_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13869952)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_self_value_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14164928)))];
            tensor<fp16, [4, 64, 384]> linear_40_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_2_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_2_attention_self_value_weight_to_fp16, x = reshape_28_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<int32, [4]> var_1253 = const()[name = tensor<string, []>("op_1253"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1254_cast_fp16 = reshape(shape = var_1253, x = linear_40_cast_fp16)[name = tensor<string, []>("op_1254_cast_fp16")];
            tensor<int32, [4]> value_layer_13_perm_0 = const()[name = tensor<string, []>("value_layer_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_43_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_43_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_43_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_43_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [4, 16, 24, 64]> transpose_76 = transpose(perm = transpose_76_perm_0, x = var_1248_cast_fp16)[name = tensor<string, []>("transpose_121")];
            tensor<fp16, [4, 16, 64, 24]> transpose_75 = transpose(perm = transpose_75_perm_0, x = var_1242_cast_fp16)[name = tensor<string, []>("transpose_122")];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_43_cast_fp16 = matmul(transpose_x = attention_scores_43_transpose_x_0, transpose_y = attention_scores_43_transpose_y_0, x = transpose_75, y = transpose_76)[name = tensor<string, []>("attention_scores_43_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_45_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_45_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [4, 16, 64, 64]> _inversed_attention_scores_45_cast_fp16 = mul(x = attention_scores_43_cast_fp16, y = _inversed_attention_scores_45_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_45_cast_fp16")];
            tensor<fp16, [1, 16, 64, 64]> var_1269_to_fp16 = const()[name = tensor<string, []>("op_1269_to_fp16"), val = tensor<fp16, [1, 16, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14165760)))];
            tensor<fp16, [4, 16, 64, 64]> input_135_cast_fp16 = add(x = _inversed_attention_scores_45_cast_fp16, y = var_1269_to_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
            tensor<fp16, [4, 16, 64, 64]> input_137_cast_fp16 = softmax(axis = var_18, x = input_135_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<bool, []> context_layer_25_transpose_x_0 = const()[name = tensor<string, []>("context_layer_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_25_transpose_y_0 = const()[name = tensor<string, []>("context_layer_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [4, 16, 64, 24]> value_layer_13_cast_fp16 = transpose(perm = value_layer_13_perm_0, x = var_1254_cast_fp16)[name = tensor<string, []>("transpose_123")];
            tensor<fp16, [4, 16, 64, 24]> context_layer_25_cast_fp16 = matmul(transpose_x = context_layer_25_transpose_x_0, transpose_y = context_layer_25_transpose_y_0, x = input_137_cast_fp16, y = value_layer_13_cast_fp16)[name = tensor<string, []>("context_layer_25_cast_fp16")];
            tensor<int32, [4]> var_1274 = const()[name = tensor<string, []>("op_1274"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1279 = const()[name = tensor<string, []>("op_1279"), val = tensor<int32, [3]>([4, 64, 384])];
            tensor<fp16, [4, 64, 16, 24]> var_1275_cast_fp16 = transpose(perm = var_1274, x = context_layer_25_cast_fp16)[name = tensor<string, []>("transpose_120")];
            tensor<fp16, [4, 64, 384]> input_139_cast_fp16 = reshape(shape = var_1279, x = var_1275_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_2_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14296896)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14591872)))];
            tensor<fp16, [4, 64, 384]> linear_41_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_2_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_2_attention_output_dense_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<int32, [4]> reshape_29_shape_0 = const()[name = tensor<string, []>("reshape_29_shape_0"), val = tensor<int32, [4]>([2, 2, 8, 3072])];
            tensor<fp16, [2, 2, 8, 3072]> reshape_29_cast_fp16 = reshape(shape = reshape_29_shape_0, x = linear_41_cast_fp16)[name = tensor<string, []>("reshape_29_cast_fp16")];
            tensor<int32, [4]> transpose_28_perm_0 = const()[name = tensor<string, []>("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_30_shape_0 = const()[name = tensor<string, []>("reshape_30_shape_0"), val = tensor<int32, [3]>([1, 256, 384])];
            tensor<fp16, [2, 8, 2, 3072]> transpose_28_cast_fp16 = transpose(perm = transpose_28_perm_0, x = reshape_29_cast_fp16)[name = tensor<string, []>("transpose_119")];
            tensor<fp16, [1, 256, 384]> reshape_30_cast_fp16 = reshape(shape = reshape_30_shape_0, x = transpose_28_cast_fp16)[name = tensor<string, []>("reshape_30_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_143_cast_fp16 = add(x = hidden_states_83_cast_fp16, y = reshape_30_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
            tensor<int32, [1]> input_145_axes_0 = const()[name = tensor<string, []>("input_145_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_layernorm_after_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14592704)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_layernorm_after_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14593536)))];
            tensor<fp16, [1, 256, 384]> input_145_cast_fp16 = layer_norm(axes = input_145_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_2_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_2_layernorm_after_weight_to_fp16, x = input_143_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<fp16, [1536, 384]> audio_model_audio_encoder_layers_2_blocks_2_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14594368)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_blocks_2_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15774080)))];
            tensor<fp16, [1, 256, 1536]> linear_42_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_2_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_2_intermediate_dense_weight_to_fp16, x = input_145_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<string, []> input_149_mode_0 = const()[name = tensor<string, []>("input_149_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 256, 1536]> input_149_cast_fp16 = gelu(mode = input_149_mode_0, x = linear_42_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")];
            tensor<fp16, [384, 1536]> audio_model_audio_encoder_layers_2_blocks_2_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15777216)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_2_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_2_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16956928)))];
            tensor<fp16, [1, 256, 384]> linear_43_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_2_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_2_output_dense_weight_to_fp16, x = input_149_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1, 256, 384]> hidden_states_93_cast_fp16 = add(x = input_143_cast_fp16, y = linear_43_cast_fp16)[name = tensor<string, []>("hidden_states_93_cast_fp16")];
            tensor<int32, [1]> hidden_states_95_axes_0 = const()[name = tensor<string, []>("hidden_states_95_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_layernorm_before_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16957760)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_layernorm_before_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16958592)))];
            tensor<fp16, [1, 256, 384]> hidden_states_95_cast_fp16 = layer_norm(axes = hidden_states_95_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_3_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_3_layernorm_before_weight_to_fp16, x = hidden_states_93_cast_fp16)[name = tensor<string, []>("hidden_states_95_cast_fp16")];
            tensor<int32, [4]> var_1329 = const()[name = tensor<string, []>("op_1329"), val = tensor<int32, [4]>([1, 16, 16, 384])];
            tensor<fp16, [1, 16, 16, 384]> input_153_cast_fp16 = reshape(shape = var_1329, x = hidden_states_95_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")];
            tensor<int32, []> gather_12_axis_0 = const()[name = tensor<string, []>("gather_12_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_12_batch_dims_0 = const()[name = tensor<string, []>("gather_12_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_12_validate_indices_0 = const()[name = tensor<string, []>("gather_12_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_21_to_uint16 = const()[name = tensor<string, []>("concat_21_to_uint16"), val = tensor<uint16, [16]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3])];
            tensor<fp16, [1, 16, 16, 384]> gather_12_cast_fp16_cast_uint16 = gather(axis = gather_12_axis_0, batch_dims = gather_12_batch_dims_0, indices = concat_21_to_uint16, validate_indices = gather_12_validate_indices_0, x = input_153_cast_fp16)[name = tensor<string, []>("gather_12_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_13_axis_0 = const()[name = tensor<string, []>("gather_13_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_13_batch_dims_0 = const()[name = tensor<string, []>("gather_13_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_13_validate_indices_0 = const()[name = tensor<string, []>("gather_13_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_22_to_uint16 = const()[name = tensor<string, []>("concat_22_to_uint16"), val = tensor<uint16, [16]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3])];
            tensor<fp16, [1, 16, 16, 384]> gather_13_cast_fp16_cast_uint16 = gather(axis = gather_13_axis_0, batch_dims = gather_13_batch_dims_0, indices = concat_22_to_uint16, validate_indices = gather_13_validate_indices_0, x = gather_12_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_13_cast_fp16_cast_uint16")];
            tensor<int32, [4]> reshape_31_shape_0 = const()[name = tensor<string, []>("reshape_31_shape_0"), val = tensor<int32, [4]>([2, 8, 2, 3072])];
            tensor<fp16, [2, 8, 2, 3072]> reshape_31_cast_fp16 = reshape(shape = reshape_31_shape_0, x = gather_13_cast_fp16_cast_uint16)[name = tensor<string, []>("reshape_31_cast_fp16")];
            tensor<int32, [4]> transpose_29_perm_0 = const()[name = tensor<string, []>("transpose_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_32_shape_0 = const()[name = tensor<string, []>("reshape_32_shape_0"), val = tensor<int32, [3]>([-1, 64, 384])];
            tensor<fp16, [2, 2, 8, 3072]> transpose_29_cast_fp16 = transpose(perm = transpose_29_perm_0, x = reshape_31_cast_fp16)[name = tensor<string, []>("transpose_118")];
            tensor<fp16, [4, 64, 384]> reshape_32_cast_fp16 = reshape(shape = reshape_32_shape_0, x = transpose_29_cast_fp16)[name = tensor<string, []>("reshape_32_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_3_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_self_query_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16959424)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_self_query_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17254400)))];
            tensor<fp16, [4, 64, 384]> linear_44_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_3_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_3_attention_self_query_weight_to_fp16, x = reshape_32_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_1430 = const()[name = tensor<string, []>("op_1430"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1431_cast_fp16 = reshape(shape = var_1430, x = linear_44_cast_fp16)[name = tensor<string, []>("op_1431_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_3_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_self_key_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17255232)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_self_key_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17550208)))];
            tensor<fp16, [4, 64, 384]> linear_45_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_3_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_3_attention_self_key_weight_to_fp16, x = reshape_32_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<int32, [4]> var_1436 = const()[name = tensor<string, []>("op_1436"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1437_cast_fp16 = reshape(shape = var_1436, x = linear_45_cast_fp16)[name = tensor<string, []>("op_1437_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_3_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_self_value_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17551040)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_self_value_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17846016)))];
            tensor<fp16, [4, 64, 384]> linear_46_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_3_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_3_attention_self_value_weight_to_fp16, x = reshape_32_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<int32, [4]> var_1442 = const()[name = tensor<string, []>("op_1442"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1443_cast_fp16 = reshape(shape = var_1442, x = linear_46_cast_fp16)[name = tensor<string, []>("op_1443_cast_fp16")];
            tensor<int32, [4]> value_layer_15_perm_0 = const()[name = tensor<string, []>("value_layer_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_47_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_47_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_47_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_47_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [4, 16, 24, 64]> transpose_78 = transpose(perm = transpose_78_perm_0, x = var_1437_cast_fp16)[name = tensor<string, []>("transpose_115")];
            tensor<fp16, [4, 16, 64, 24]> transpose_77 = transpose(perm = transpose_77_perm_0, x = var_1431_cast_fp16)[name = tensor<string, []>("transpose_116")];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_47_cast_fp16 = matmul(transpose_x = attention_scores_47_transpose_x_0, transpose_y = attention_scores_47_transpose_y_0, x = transpose_77, y = transpose_78)[name = tensor<string, []>("attention_scores_47_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_49_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_49_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [4, 16, 64, 64]> _inversed_attention_scores_49_cast_fp16 = mul(x = attention_scores_47_cast_fp16, y = _inversed_attention_scores_49_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_49_cast_fp16")];
            tensor<fp16, [1, 16, 64, 64]> var_1458_to_fp16 = const()[name = tensor<string, []>("op_1458_to_fp16"), val = tensor<fp16, [1, 16, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17846848)))];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_51_cast_fp16 = add(x = _inversed_attention_scores_49_cast_fp16, y = var_1458_to_fp16)[name = tensor<string, []>("attention_scores_51_cast_fp16")];
            tensor<int32, [5]> var_1464 = const()[name = tensor<string, []>("op_1464"), val = tensor<int32, [5]>([1, 4, 16, 64, 64])];
            tensor<fp16, [1, 4, 16, 64, 64]> attention_scores_53_cast_fp16 = reshape(shape = var_1464, x = attention_scores_51_cast_fp16)[name = tensor<string, []>("attention_scores_53_cast_fp16")];
            tensor<fp16, [1, 4, 16, 64, 64]> attention_scores_55_cast_fp16 = add(x = attention_scores_53_cast_fp16, y = var_1137_to_fp16)[name = tensor<string, []>("attention_scores_55_cast_fp16")];
            tensor<int32, [4]> var_1469 = const()[name = tensor<string, []>("op_1469"), val = tensor<int32, [4]>([-1, 16, 64, 64])];
            tensor<fp16, [4, 16, 64, 64]> input_155_cast_fp16 = reshape(shape = var_1469, x = attention_scores_55_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
            tensor<fp16, [4, 16, 64, 64]> input_157_cast_fp16 = softmax(axis = var_18, x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<bool, []> context_layer_29_transpose_x_0 = const()[name = tensor<string, []>("context_layer_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_29_transpose_y_0 = const()[name = tensor<string, []>("context_layer_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [4, 16, 64, 24]> value_layer_15_cast_fp16 = transpose(perm = value_layer_15_perm_0, x = var_1443_cast_fp16)[name = tensor<string, []>("transpose_117")];
            tensor<fp16, [4, 16, 64, 24]> context_layer_29_cast_fp16 = matmul(transpose_x = context_layer_29_transpose_x_0, transpose_y = context_layer_29_transpose_y_0, x = input_157_cast_fp16, y = value_layer_15_cast_fp16)[name = tensor<string, []>("context_layer_29_cast_fp16")];
            tensor<int32, [4]> var_1474 = const()[name = tensor<string, []>("op_1474"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1479 = const()[name = tensor<string, []>("op_1479"), val = tensor<int32, [3]>([4, 64, 384])];
            tensor<fp16, [4, 64, 16, 24]> var_1475_cast_fp16 = transpose(perm = var_1474, x = context_layer_29_cast_fp16)[name = tensor<string, []>("transpose_114")];
            tensor<fp16, [4, 64, 384]> input_159_cast_fp16 = reshape(shape = var_1479, x = var_1475_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_3_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17977984)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18272960)))];
            tensor<fp16, [4, 64, 384]> linear_47_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_3_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_3_attention_output_dense_weight_to_fp16, x = input_159_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<int32, [4]> reshape_33_shape_0 = const()[name = tensor<string, []>("reshape_33_shape_0"), val = tensor<int32, [4]>([2, 2, 8, 3072])];
            tensor<fp16, [2, 2, 8, 3072]> reshape_33_cast_fp16 = reshape(shape = reshape_33_shape_0, x = linear_47_cast_fp16)[name = tensor<string, []>("reshape_33_cast_fp16")];
            tensor<int32, [4]> transpose_30_perm_0 = const()[name = tensor<string, []>("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> reshape_34_shape_0 = const()[name = tensor<string, []>("reshape_34_shape_0"), val = tensor<int32, [4]>([-1, 16, 16, 384])];
            tensor<fp16, [2, 8, 2, 3072]> transpose_30_cast_fp16 = transpose(perm = transpose_30_perm_0, x = reshape_33_cast_fp16)[name = tensor<string, []>("transpose_113")];
            tensor<fp16, [1, 16, 16, 384]> reshape_34_cast_fp16 = reshape(shape = reshape_34_shape_0, x = transpose_30_cast_fp16)[name = tensor<string, []>("reshape_34_cast_fp16")];
            tensor<int32, []> gather_14_axis_0 = const()[name = tensor<string, []>("gather_14_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_14_batch_dims_0 = const()[name = tensor<string, []>("gather_14_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_14_validate_indices_0 = const()[name = tensor<string, []>("gather_14_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_23_to_uint16 = const()[name = tensor<string, []>("concat_23_to_uint16"), val = tensor<uint16, [16]>([12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])];
            tensor<fp16, [1, 16, 16, 384]> gather_14_cast_fp16_cast_uint16 = gather(axis = gather_14_axis_0, batch_dims = gather_14_batch_dims_0, indices = concat_23_to_uint16, validate_indices = gather_14_validate_indices_0, x = reshape_34_cast_fp16)[name = tensor<string, []>("gather_14_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_15_axis_0 = const()[name = tensor<string, []>("gather_15_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_15_batch_dims_0 = const()[name = tensor<string, []>("gather_15_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_15_validate_indices_0 = const()[name = tensor<string, []>("gather_15_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_24_to_uint16 = const()[name = tensor<string, []>("concat_24_to_uint16"), val = tensor<uint16, [16]>([12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])];
            tensor<fp16, [1, 16, 16, 384]> gather_15_cast_fp16_cast_uint16 = gather(axis = gather_15_axis_0, batch_dims = gather_15_batch_dims_0, indices = concat_24_to_uint16, validate_indices = gather_15_validate_indices_0, x = gather_14_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_15_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_1503 = const()[name = tensor<string, []>("op_1503"), val = tensor<int32, [3]>([1, 256, 384])];
            tensor<fp16, [1, 256, 384]> attention_windows_31_cast_fp16 = reshape(shape = var_1503, x = gather_15_cast_fp16_cast_uint16)[name = tensor<string, []>("attention_windows_31_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_163_cast_fp16 = add(x = hidden_states_93_cast_fp16, y = attention_windows_31_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
            tensor<int32, [1]> input_165_axes_0 = const()[name = tensor<string, []>("input_165_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_layernorm_after_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18273792)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_layernorm_after_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18274624)))];
            tensor<fp16, [1, 256, 384]> input_165_cast_fp16 = layer_norm(axes = input_165_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_3_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_3_layernorm_after_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")];
            tensor<fp16, [1536, 384]> audio_model_audio_encoder_layers_2_blocks_3_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18275456)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_blocks_3_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19455168)))];
            tensor<fp16, [1, 256, 1536]> linear_48_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_3_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_3_intermediate_dense_weight_to_fp16, x = input_165_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<string, []> input_169_mode_0 = const()[name = tensor<string, []>("input_169_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 256, 1536]> input_169_cast_fp16 = gelu(mode = input_169_mode_0, x = linear_48_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")];
            tensor<fp16, [384, 1536]> audio_model_audio_encoder_layers_2_blocks_3_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19458304)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_3_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_3_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20638016)))];
            tensor<fp16, [1, 256, 384]> linear_49_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_3_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_3_output_dense_weight_to_fp16, x = input_169_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, [1, 256, 384]> hidden_states_109_cast_fp16 = add(x = input_163_cast_fp16, y = linear_49_cast_fp16)[name = tensor<string, []>("hidden_states_109_cast_fp16")];
            tensor<int32, [1]> hidden_states_111_axes_0 = const()[name = tensor<string, []>("hidden_states_111_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_layernorm_before_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20638848)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_layernorm_before_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20639680)))];
            tensor<fp16, [1, 256, 384]> hidden_states_111_cast_fp16 = layer_norm(axes = hidden_states_111_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_4_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_4_layernorm_before_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("hidden_states_111_cast_fp16")];
            tensor<int32, [4]> reshape_35_shape_0 = const()[name = tensor<string, []>("reshape_35_shape_0"), val = tensor<int32, [4]>([2, 8, 2, 3072])];
            tensor<fp16, [2, 8, 2, 3072]> reshape_35_cast_fp16 = reshape(shape = reshape_35_shape_0, x = hidden_states_111_cast_fp16)[name = tensor<string, []>("reshape_35_cast_fp16")];
            tensor<int32, [4]> transpose_31_perm_0 = const()[name = tensor<string, []>("transpose_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_36_shape_0 = const()[name = tensor<string, []>("reshape_36_shape_0"), val = tensor<int32, [3]>([-1, 64, 384])];
            tensor<fp16, [2, 2, 8, 3072]> transpose_31_cast_fp16 = transpose(perm = transpose_31_perm_0, x = reshape_35_cast_fp16)[name = tensor<string, []>("transpose_112")];
            tensor<fp16, [4, 64, 384]> reshape_36_cast_fp16 = reshape(shape = reshape_36_shape_0, x = transpose_31_cast_fp16)[name = tensor<string, []>("reshape_36_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_4_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_self_query_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20640512)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_self_query_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20935488)))];
            tensor<fp16, [4, 64, 384]> linear_50_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_4_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_4_attention_self_query_weight_to_fp16, x = reshape_36_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_1571 = const()[name = tensor<string, []>("op_1571"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1572_cast_fp16 = reshape(shape = var_1571, x = linear_50_cast_fp16)[name = tensor<string, []>("op_1572_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_4_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_self_key_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20936320)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_self_key_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21231296)))];
            tensor<fp16, [4, 64, 384]> linear_51_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_4_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_4_attention_self_key_weight_to_fp16, x = reshape_36_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<int32, [4]> var_1577 = const()[name = tensor<string, []>("op_1577"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1578_cast_fp16 = reshape(shape = var_1577, x = linear_51_cast_fp16)[name = tensor<string, []>("op_1578_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_4_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_self_value_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21232128)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_self_value_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21527104)))];
            tensor<fp16, [4, 64, 384]> linear_52_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_4_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_4_attention_self_value_weight_to_fp16, x = reshape_36_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<int32, [4]> var_1583 = const()[name = tensor<string, []>("op_1583"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1584_cast_fp16 = reshape(shape = var_1583, x = linear_52_cast_fp16)[name = tensor<string, []>("op_1584_cast_fp16")];
            tensor<int32, [4]> value_layer_17_perm_0 = const()[name = tensor<string, []>("value_layer_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_57_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_57_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [4, 16, 24, 64]> transpose_80 = transpose(perm = transpose_80_perm_0, x = var_1578_cast_fp16)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [4, 16, 64, 24]> transpose_79 = transpose(perm = transpose_79_perm_0, x = var_1572_cast_fp16)[name = tensor<string, []>("transpose_110")];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_57_cast_fp16 = matmul(transpose_x = attention_scores_57_transpose_x_0, transpose_y = attention_scores_57_transpose_y_0, x = transpose_79, y = transpose_80)[name = tensor<string, []>("attention_scores_57_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_59_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_59_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [4, 16, 64, 64]> _inversed_attention_scores_59_cast_fp16 = mul(x = attention_scores_57_cast_fp16, y = _inversed_attention_scores_59_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_59_cast_fp16")];
            tensor<fp16, [1, 16, 64, 64]> var_1599_to_fp16 = const()[name = tensor<string, []>("op_1599_to_fp16"), val = tensor<fp16, [1, 16, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21527936)))];
            tensor<fp16, [4, 16, 64, 64]> input_175_cast_fp16 = add(x = _inversed_attention_scores_59_cast_fp16, y = var_1599_to_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<fp16, [4, 16, 64, 64]> input_177_cast_fp16 = softmax(axis = var_18, x = input_175_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")];
            tensor<bool, []> context_layer_33_transpose_x_0 = const()[name = tensor<string, []>("context_layer_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_33_transpose_y_0 = const()[name = tensor<string, []>("context_layer_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [4, 16, 64, 24]> value_layer_17_cast_fp16 = transpose(perm = value_layer_17_perm_0, x = var_1584_cast_fp16)[name = tensor<string, []>("transpose_111")];
            tensor<fp16, [4, 16, 64, 24]> context_layer_33_cast_fp16 = matmul(transpose_x = context_layer_33_transpose_x_0, transpose_y = context_layer_33_transpose_y_0, x = input_177_cast_fp16, y = value_layer_17_cast_fp16)[name = tensor<string, []>("context_layer_33_cast_fp16")];
            tensor<int32, [4]> var_1604 = const()[name = tensor<string, []>("op_1604"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1609 = const()[name = tensor<string, []>("op_1609"), val = tensor<int32, [3]>([4, 64, 384])];
            tensor<fp16, [4, 64, 16, 24]> var_1605_cast_fp16 = transpose(perm = var_1604, x = context_layer_33_cast_fp16)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [4, 64, 384]> input_179_cast_fp16 = reshape(shape = var_1609, x = var_1605_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_4_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21659072)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21954048)))];
            tensor<fp16, [4, 64, 384]> linear_53_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_4_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_4_attention_output_dense_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<int32, [4]> reshape_37_shape_0 = const()[name = tensor<string, []>("reshape_37_shape_0"), val = tensor<int32, [4]>([2, 2, 8, 3072])];
            tensor<fp16, [2, 2, 8, 3072]> reshape_37_cast_fp16 = reshape(shape = reshape_37_shape_0, x = linear_53_cast_fp16)[name = tensor<string, []>("reshape_37_cast_fp16")];
            tensor<int32, [4]> transpose_32_perm_0 = const()[name = tensor<string, []>("transpose_32_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_38_shape_0 = const()[name = tensor<string, []>("reshape_38_shape_0"), val = tensor<int32, [3]>([1, 256, 384])];
            tensor<fp16, [2, 8, 2, 3072]> transpose_32_cast_fp16 = transpose(perm = transpose_32_perm_0, x = reshape_37_cast_fp16)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 256, 384]> reshape_38_cast_fp16 = reshape(shape = reshape_38_shape_0, x = transpose_32_cast_fp16)[name = tensor<string, []>("reshape_38_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_183_cast_fp16 = add(x = hidden_states_109_cast_fp16, y = reshape_38_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")];
            tensor<int32, [1]> input_185_axes_0 = const()[name = tensor<string, []>("input_185_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_layernorm_after_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21954880)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_layernorm_after_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21955712)))];
            tensor<fp16, [1, 256, 384]> input_185_cast_fp16 = layer_norm(axes = input_185_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_4_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_4_layernorm_after_weight_to_fp16, x = input_183_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")];
            tensor<fp16, [1536, 384]> audio_model_audio_encoder_layers_2_blocks_4_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21956544)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_blocks_4_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23136256)))];
            tensor<fp16, [1, 256, 1536]> linear_54_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_4_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_4_intermediate_dense_weight_to_fp16, x = input_185_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<string, []> input_189_mode_0 = const()[name = tensor<string, []>("input_189_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 256, 1536]> input_189_cast_fp16 = gelu(mode = input_189_mode_0, x = linear_54_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")];
            tensor<fp16, [384, 1536]> audio_model_audio_encoder_layers_2_blocks_4_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23139392)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_4_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_4_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24319104)))];
            tensor<fp16, [1, 256, 384]> linear_55_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_4_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_4_output_dense_weight_to_fp16, x = input_189_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, [1, 256, 384]> hidden_states_119_cast_fp16 = add(x = input_183_cast_fp16, y = linear_55_cast_fp16)[name = tensor<string, []>("hidden_states_119_cast_fp16")];
            tensor<int32, [1]> hidden_states_121_axes_0 = const()[name = tensor<string, []>("hidden_states_121_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_layernorm_before_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24319936)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_layernorm_before_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24320768)))];
            tensor<fp16, [1, 256, 384]> hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_5_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_5_layernorm_before_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor<string, []>("hidden_states_121_cast_fp16")];
            tensor<int32, [4]> var_1659 = const()[name = tensor<string, []>("op_1659"), val = tensor<int32, [4]>([1, 16, 16, 384])];
            tensor<fp16, [1, 16, 16, 384]> input_193_cast_fp16 = reshape(shape = var_1659, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")];
            tensor<int32, []> gather_16_axis_0 = const()[name = tensor<string, []>("gather_16_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_16_batch_dims_0 = const()[name = tensor<string, []>("gather_16_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_16_validate_indices_0 = const()[name = tensor<string, []>("gather_16_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_25_to_uint16 = const()[name = tensor<string, []>("concat_25_to_uint16"), val = tensor<uint16, [16]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3])];
            tensor<fp16, [1, 16, 16, 384]> gather_16_cast_fp16_cast_uint16 = gather(axis = gather_16_axis_0, batch_dims = gather_16_batch_dims_0, indices = concat_25_to_uint16, validate_indices = gather_16_validate_indices_0, x = input_193_cast_fp16)[name = tensor<string, []>("gather_16_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_17_axis_0 = const()[name = tensor<string, []>("gather_17_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_17_batch_dims_0 = const()[name = tensor<string, []>("gather_17_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_17_validate_indices_0 = const()[name = tensor<string, []>("gather_17_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_26_to_uint16 = const()[name = tensor<string, []>("concat_26_to_uint16"), val = tensor<uint16, [16]>([4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3])];
            tensor<fp16, [1, 16, 16, 384]> gather_17_cast_fp16_cast_uint16 = gather(axis = gather_17_axis_0, batch_dims = gather_17_batch_dims_0, indices = concat_26_to_uint16, validate_indices = gather_17_validate_indices_0, x = gather_16_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_17_cast_fp16_cast_uint16")];
            tensor<int32, [4]> reshape_39_shape_0 = const()[name = tensor<string, []>("reshape_39_shape_0"), val = tensor<int32, [4]>([2, 8, 2, 3072])];
            tensor<fp16, [2, 8, 2, 3072]> reshape_39_cast_fp16 = reshape(shape = reshape_39_shape_0, x = gather_17_cast_fp16_cast_uint16)[name = tensor<string, []>("reshape_39_cast_fp16")];
            tensor<int32, [4]> transpose_33_perm_0 = const()[name = tensor<string, []>("transpose_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_40_shape_0 = const()[name = tensor<string, []>("reshape_40_shape_0"), val = tensor<int32, [3]>([-1, 64, 384])];
            tensor<fp16, [2, 2, 8, 3072]> transpose_33_cast_fp16 = transpose(perm = transpose_33_perm_0, x = reshape_39_cast_fp16)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [4, 64, 384]> reshape_40_cast_fp16 = reshape(shape = reshape_40_shape_0, x = transpose_33_cast_fp16)[name = tensor<string, []>("reshape_40_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_5_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_self_query_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24321600)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_self_query_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24616576)))];
            tensor<fp16, [4, 64, 384]> linear_56_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_5_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_5_attention_self_query_weight_to_fp16, x = reshape_40_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_1760 = const()[name = tensor<string, []>("op_1760"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1761_cast_fp16 = reshape(shape = var_1760, x = linear_56_cast_fp16)[name = tensor<string, []>("op_1761_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_5_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_self_key_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24617408)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_self_key_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24912384)))];
            tensor<fp16, [4, 64, 384]> linear_57_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_5_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_5_attention_self_key_weight_to_fp16, x = reshape_40_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<int32, [4]> var_1766 = const()[name = tensor<string, []>("op_1766"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1767_cast_fp16 = reshape(shape = var_1766, x = linear_57_cast_fp16)[name = tensor<string, []>("op_1767_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_5_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_self_value_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24913216)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_self_value_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25208192)))];
            tensor<fp16, [4, 64, 384]> linear_58_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_5_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_5_attention_self_value_weight_to_fp16, x = reshape_40_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<int32, [4]> var_1772 = const()[name = tensor<string, []>("op_1772"), val = tensor<int32, [4]>([4, 64, -1, 24])];
            tensor<fp16, [4, 64, 16, 24]> var_1773_cast_fp16 = reshape(shape = var_1772, x = linear_58_cast_fp16)[name = tensor<string, []>("op_1773_cast_fp16")];
            tensor<int32, [4]> value_layer_19_perm_0 = const()[name = tensor<string, []>("value_layer_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_61_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_61_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_61_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_61_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [4, 16, 24, 64]> transpose_82 = transpose(perm = transpose_82_perm_0, x = var_1767_cast_fp16)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [4, 16, 64, 24]> transpose_81 = transpose(perm = transpose_81_perm_0, x = var_1761_cast_fp16)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_61_cast_fp16 = matmul(transpose_x = attention_scores_61_transpose_x_0, transpose_y = attention_scores_61_transpose_y_0, x = transpose_81, y = transpose_82)[name = tensor<string, []>("attention_scores_61_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_63_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_63_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [4, 16, 64, 64]> _inversed_attention_scores_63_cast_fp16 = mul(x = attention_scores_61_cast_fp16, y = _inversed_attention_scores_63_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_63_cast_fp16")];
            tensor<fp16, [1, 16, 64, 64]> var_1788_to_fp16 = const()[name = tensor<string, []>("op_1788_to_fp16"), val = tensor<fp16, [1, 16, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25209024)))];
            tensor<fp16, [4, 16, 64, 64]> attention_scores_65_cast_fp16 = add(x = _inversed_attention_scores_63_cast_fp16, y = var_1788_to_fp16)[name = tensor<string, []>("attention_scores_65_cast_fp16")];
            tensor<int32, [5]> var_1794 = const()[name = tensor<string, []>("op_1794"), val = tensor<int32, [5]>([1, 4, 16, 64, 64])];
            tensor<fp16, [1, 4, 16, 64, 64]> attention_scores_67_cast_fp16 = reshape(shape = var_1794, x = attention_scores_65_cast_fp16)[name = tensor<string, []>("attention_scores_67_cast_fp16")];
            tensor<fp16, [1, 4, 16, 64, 64]> attention_scores_69_cast_fp16 = add(x = attention_scores_67_cast_fp16, y = var_1137_to_fp16)[name = tensor<string, []>("attention_scores_69_cast_fp16")];
            tensor<int32, [4]> var_1799 = const()[name = tensor<string, []>("op_1799"), val = tensor<int32, [4]>([-1, 16, 64, 64])];
            tensor<fp16, [4, 16, 64, 64]> input_195_cast_fp16 = reshape(shape = var_1799, x = attention_scores_69_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
            tensor<fp16, [4, 16, 64, 64]> input_197_cast_fp16 = softmax(axis = var_18, x = input_195_cast_fp16)[name = tensor<string, []>("input_197_cast_fp16")];
            tensor<bool, []> context_layer_37_transpose_x_0 = const()[name = tensor<string, []>("context_layer_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_37_transpose_y_0 = const()[name = tensor<string, []>("context_layer_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [4, 16, 64, 24]> value_layer_19_cast_fp16 = transpose(perm = value_layer_19_perm_0, x = var_1773_cast_fp16)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [4, 16, 64, 24]> context_layer_37_cast_fp16 = matmul(transpose_x = context_layer_37_transpose_x_0, transpose_y = context_layer_37_transpose_y_0, x = input_197_cast_fp16, y = value_layer_19_cast_fp16)[name = tensor<string, []>("context_layer_37_cast_fp16")];
            tensor<int32, [4]> var_1804 = const()[name = tensor<string, []>("op_1804"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1809 = const()[name = tensor<string, []>("op_1809"), val = tensor<int32, [3]>([4, 64, 384])];
            tensor<fp16, [4, 64, 16, 24]> var_1805_cast_fp16 = transpose(perm = var_1804, x = context_layer_37_cast_fp16)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [4, 64, 384]> input_199_cast_fp16 = reshape(shape = var_1809, x = var_1805_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")];
            tensor<fp16, [384, 384]> audio_model_audio_encoder_layers_2_blocks_5_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25340160)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25635136)))];
            tensor<fp16, [4, 64, 384]> linear_59_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_5_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_5_attention_output_dense_weight_to_fp16, x = input_199_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<int32, [4]> reshape_41_shape_0 = const()[name = tensor<string, []>("reshape_41_shape_0"), val = tensor<int32, [4]>([2, 2, 8, 3072])];
            tensor<fp16, [2, 2, 8, 3072]> reshape_41_cast_fp16 = reshape(shape = reshape_41_shape_0, x = linear_59_cast_fp16)[name = tensor<string, []>("reshape_41_cast_fp16")];
            tensor<int32, [4]> transpose_34_perm_0 = const()[name = tensor<string, []>("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> reshape_42_shape_0 = const()[name = tensor<string, []>("reshape_42_shape_0"), val = tensor<int32, [4]>([-1, 16, 16, 384])];
            tensor<fp16, [2, 8, 2, 3072]> transpose_34_cast_fp16 = transpose(perm = transpose_34_perm_0, x = reshape_41_cast_fp16)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [1, 16, 16, 384]> reshape_42_cast_fp16 = reshape(shape = reshape_42_shape_0, x = transpose_34_cast_fp16)[name = tensor<string, []>("reshape_42_cast_fp16")];
            tensor<int32, []> gather_18_axis_0 = const()[name = tensor<string, []>("gather_18_axis_0"), val = tensor<int32, []>(1)];
            tensor<int32, []> gather_18_batch_dims_0 = const()[name = tensor<string, []>("gather_18_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_18_validate_indices_0 = const()[name = tensor<string, []>("gather_18_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_27_to_uint16 = const()[name = tensor<string, []>("concat_27_to_uint16"), val = tensor<uint16, [16]>([12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])];
            tensor<fp16, [1, 16, 16, 384]> gather_18_cast_fp16_cast_uint16 = gather(axis = gather_18_axis_0, batch_dims = gather_18_batch_dims_0, indices = concat_27_to_uint16, validate_indices = gather_18_validate_indices_0, x = reshape_42_cast_fp16)[name = tensor<string, []>("gather_18_cast_fp16_cast_uint16")];
            tensor<int32, []> gather_19_axis_0 = const()[name = tensor<string, []>("gather_19_axis_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> gather_19_batch_dims_0 = const()[name = tensor<string, []>("gather_19_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> gather_19_validate_indices_0 = const()[name = tensor<string, []>("gather_19_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<uint16, [16]> concat_28_to_uint16 = const()[name = tensor<string, []>("concat_28_to_uint16"), val = tensor<uint16, [16]>([12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])];
            tensor<fp16, [1, 16, 16, 384]> gather_19_cast_fp16_cast_uint16 = gather(axis = gather_19_axis_0, batch_dims = gather_19_batch_dims_0, indices = concat_28_to_uint16, validate_indices = gather_19_validate_indices_0, x = gather_18_cast_fp16_cast_uint16)[name = tensor<string, []>("gather_19_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_1833 = const()[name = tensor<string, []>("op_1833"), val = tensor<int32, [3]>([1, 256, 384])];
            tensor<fp16, [1, 256, 384]> attention_windows_39_cast_fp16 = reshape(shape = var_1833, x = gather_19_cast_fp16_cast_uint16)[name = tensor<string, []>("attention_windows_39_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_203_cast_fp16 = add(x = hidden_states_119_cast_fp16, y = attention_windows_39_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")];
            tensor<int32, [1]> input_205_axes_0 = const()[name = tensor<string, []>("input_205_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_layernorm_after_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25635968)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_layernorm_after_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25636800)))];
            tensor<fp16, [1, 256, 384]> input_205_cast_fp16 = layer_norm(axes = input_205_axes_0, beta = audio_model_audio_encoder_layers_2_blocks_5_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_blocks_5_layernorm_after_weight_to_fp16, x = input_203_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")];
            tensor<fp16, [1536, 384]> audio_model_audio_encoder_layers_2_blocks_5_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25637632)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_blocks_5_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26817344)))];
            tensor<fp16, [1, 256, 1536]> linear_60_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_5_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_5_intermediate_dense_weight_to_fp16, x = input_205_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<string, []> input_209_mode_0 = const()[name = tensor<string, []>("input_209_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 256, 1536]> input_209_cast_fp16 = gelu(mode = input_209_mode_0, x = linear_60_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")];
            tensor<fp16, [384, 1536]> audio_model_audio_encoder_layers_2_blocks_5_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_output_dense_weight_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26820480)))];
            tensor<fp16, [384]> audio_model_audio_encoder_layers_2_blocks_5_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_blocks_5_output_dense_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28000192)))];
            tensor<fp16, [1, 256, 384]> linear_61_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_2_blocks_5_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_2_blocks_5_output_dense_weight_to_fp16, x = input_209_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1, 256, 384]> input_feature_13_cast_fp16 = add(x = input_203_cast_fp16, y = linear_61_cast_fp16)[name = tensor<string, []>("input_feature_13_cast_fp16")];
            tensor<int32, [4]> var_1856 = const()[name = tensor<string, []>("op_1856"), val = tensor<int32, [4]>([1, 16, 16, 384])];
            tensor<fp16, [1, 16, 16, 384]> input_feature_15_cast_fp16 = reshape(shape = var_1856, x = input_feature_13_cast_fp16)[name = tensor<string, []>("input_feature_15_cast_fp16")];
            tensor<int32, [4]> var_1859_begin_0 = const()[name = tensor<string, []>("op_1859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1859_end_0 = const()[name = tensor<string, []>("op_1859_end_0"), val = tensor<int32, [4]>([1, 16, 16, 384])];
            tensor<bool, [4]> var_1859_end_mask_0 = const()[name = tensor<string, []>("op_1859_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_1859_stride_0 = const()[name = tensor<string, []>("op_1859_stride_0"), val = tensor<int32, [4]>([1, 2, 1, 1])];
            tensor<fp16, [1, 8, 16, 384]> var_1859_cast_fp16 = slice_by_index(begin = var_1859_begin_0, end = var_1859_end_0, end_mask = var_1859_end_mask_0, stride = var_1859_stride_0, x = input_feature_15_cast_fp16)[name = tensor<string, []>("op_1859_cast_fp16")];
            tensor<int32, [4]> var_1860_begin_0 = const()[name = tensor<string, []>("op_1860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1860_end_0 = const()[name = tensor<string, []>("op_1860_end_0"), val = tensor<int32, [4]>([1, 8, 16, 384])];
            tensor<bool, [4]> var_1860_end_mask_0 = const()[name = tensor<string, []>("op_1860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_1860_stride_0 = const()[name = tensor<string, []>("op_1860_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 8, 8, 384]> var_1860_cast_fp16 = slice_by_index(begin = var_1860_begin_0, end = var_1860_end_0, end_mask = var_1860_end_mask_0, stride = var_1860_stride_0, x = var_1859_cast_fp16)[name = tensor<string, []>("op_1860_cast_fp16")];
            tensor<int32, [4]> var_1863_begin_0 = const()[name = tensor<string, []>("op_1863_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])];
            tensor<int32, [4]> var_1863_end_0 = const()[name = tensor<string, []>("op_1863_end_0"), val = tensor<int32, [4]>([1, 16, 16, 384])];
            tensor<bool, [4]> var_1863_end_mask_0 = const()[name = tensor<string, []>("op_1863_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_1863_stride_0 = const()[name = tensor<string, []>("op_1863_stride_0"), val = tensor<int32, [4]>([1, 2, 1, 1])];
            tensor<fp16, [1, 8, 16, 384]> var_1863_cast_fp16 = slice_by_index(begin = var_1863_begin_0, end = var_1863_end_0, end_mask = var_1863_end_mask_0, stride = var_1863_stride_0, x = input_feature_15_cast_fp16)[name = tensor<string, []>("op_1863_cast_fp16")];
            tensor<int32, [4]> var_1864_begin_0 = const()[name = tensor<string, []>("op_1864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1864_end_0 = const()[name = tensor<string, []>("op_1864_end_0"), val = tensor<int32, [4]>([1, 8, 16, 384])];
            tensor<bool, [4]> var_1864_end_mask_0 = const()[name = tensor<string, []>("op_1864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_1864_stride_0 = const()[name = tensor<string, []>("op_1864_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 8, 8, 384]> var_1864_cast_fp16 = slice_by_index(begin = var_1864_begin_0, end = var_1864_end_0, end_mask = var_1864_end_mask_0, stride = var_1864_stride_0, x = var_1863_cast_fp16)[name = tensor<string, []>("op_1864_cast_fp16")];
            tensor<int32, [4]> var_1868_begin_0 = const()[name = tensor<string, []>("op_1868_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1868_end_0 = const()[name = tensor<string, []>("op_1868_end_0"), val = tensor<int32, [4]>([1, 8, 16, 384])];
            tensor<bool, [4]> var_1868_end_mask_0 = const()[name = tensor<string, []>("op_1868_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_1868_stride_0 = const()[name = tensor<string, []>("op_1868_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 8, 8, 384]> var_1868_cast_fp16 = slice_by_index(begin = var_1868_begin_0, end = var_1868_end_0, end_mask = var_1868_end_mask_0, stride = var_1868_stride_0, x = var_1859_cast_fp16)[name = tensor<string, []>("op_1868_cast_fp16")];
            tensor<int32, [4]> var_1872_begin_0 = const()[name = tensor<string, []>("op_1872_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1872_end_0 = const()[name = tensor<string, []>("op_1872_end_0"), val = tensor<int32, [4]>([1, 8, 16, 384])];
            tensor<bool, [4]> var_1872_end_mask_0 = const()[name = tensor<string, []>("op_1872_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<int32, [4]> var_1872_stride_0 = const()[name = tensor<string, []>("op_1872_stride_0"), val = tensor<int32, [4]>([1, 1, 2, 1])];
            tensor<fp16, [1, 8, 8, 384]> var_1872_cast_fp16 = slice_by_index(begin = var_1872_begin_0, end = var_1872_end_0, end_mask = var_1872_end_mask_0, stride = var_1872_stride_0, x = var_1863_cast_fp16)[name = tensor<string, []>("op_1872_cast_fp16")];
            tensor<bool, []> input_feature_interleave_0 = const()[name = tensor<string, []>("input_feature_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 8, 1536]> input_feature_cast_fp16 = concat(axis = var_18, interleave = input_feature_interleave_0, values = (var_1860_cast_fp16, var_1864_cast_fp16, var_1868_cast_fp16, var_1872_cast_fp16))[name = tensor<string, []>("input_feature_cast_fp16")];
            tensor<int32, [3]> var_1879 = const()[name = tensor<string, []>("op_1879"), val = tensor<int32, [3]>([1, -1, 1536])];
            tensor<fp16, [1, 64, 1536]> input_213_cast_fp16 = reshape(shape = var_1879, x = input_feature_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")];
            tensor<int32, [1]> input_215_axes_0 = const()[name = tensor<string, []>("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_downsample_norm_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_downsample_norm_weight_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28001024)))];
            tensor<fp16, [1536]> audio_model_audio_encoder_layers_2_downsample_norm_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_downsample_norm_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28004160)))];
            tensor<fp16, [1, 64, 1536]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = audio_model_audio_encoder_layers_2_downsample_norm_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_2_downsample_norm_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
            tensor<fp16, [768, 1536]> audio_model_audio_encoder_layers_2_downsample_reduction_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_2_downsample_reduction_weight_to_fp16"), val = tensor<fp16, [768, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28007296)))];
            tensor<fp16, [768]> linear_62_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_62_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30366656)))];
            tensor<fp16, [1, 64, 768]> linear_62_cast_fp16 = linear(bias = linear_62_bias_0_to_fp16, weight = audio_model_audio_encoder_layers_2_downsample_reduction_weight_to_fp16, x = input_215_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [1]> hidden_states_137_axes_0 = const()[name = tensor<string, []>("hidden_states_137_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_layernorm_before_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30368256)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_layernorm_before_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30369856)))];
            tensor<fp16, [1, 64, 768]> hidden_states_137_cast_fp16 = layer_norm(axes = hidden_states_137_axes_0, beta = audio_model_audio_encoder_layers_3_blocks_0_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_3_blocks_0_layernorm_before_weight_to_fp16, x = linear_62_cast_fp16)[name = tensor<string, []>("hidden_states_137_cast_fp16")];
            tensor<int32, [4]> reshape_43_shape_0 = const()[name = tensor<string, []>("reshape_43_shape_0"), val = tensor<int32, [4]>([1, 8, 1, 6144])];
            tensor<fp16, [1, 8, 1, 6144]> reshape_43_cast_fp16 = reshape(shape = reshape_43_shape_0, x = hidden_states_137_cast_fp16)[name = tensor<string, []>("reshape_43_cast_fp16")];
            tensor<int32, [4]> transpose_35_perm_0 = const()[name = tensor<string, []>("transpose_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_44_shape_0 = const()[name = tensor<string, []>("reshape_44_shape_0"), val = tensor<int32, [3]>([-1, 64, 768])];
            tensor<fp16, [1, 1, 8, 6144]> transpose_35_cast_fp16 = transpose(perm = transpose_35_perm_0, x = reshape_43_cast_fp16)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [1, 64, 768]> reshape_44_cast_fp16 = reshape(shape = reshape_44_shape_0, x = transpose_35_cast_fp16)[name = tensor<string, []>("reshape_44_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_0_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30371456)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31551168)))];
            tensor<fp16, [1, 64, 768]> linear_63_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_0_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_0_attention_self_query_weight_to_fp16, x = reshape_44_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<int32, [4]> var_1962 = const()[name = tensor<string, []>("op_1962"), val = tensor<int32, [4]>([1, 64, -1, 24])];
            tensor<fp16, [1, 64, 32, 24]> var_1963_cast_fp16 = reshape(shape = var_1962, x = linear_63_cast_fp16)[name = tensor<string, []>("op_1963_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_0_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31552768)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32732480)))];
            tensor<fp16, [1, 64, 768]> linear_64_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_0_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_0_attention_self_key_weight_to_fp16, x = reshape_44_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<int32, [4]> var_1968 = const()[name = tensor<string, []>("op_1968"), val = tensor<int32, [4]>([1, 64, -1, 24])];
            tensor<fp16, [1, 64, 32, 24]> var_1969_cast_fp16 = reshape(shape = var_1968, x = linear_64_cast_fp16)[name = tensor<string, []>("op_1969_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_0_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32734080)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33913792)))];
            tensor<fp16, [1, 64, 768]> linear_65_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_0_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_0_attention_self_value_weight_to_fp16, x = reshape_44_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<int32, [4]> var_1974 = const()[name = tensor<string, []>("op_1974"), val = tensor<int32, [4]>([1, 64, -1, 24])];
            tensor<fp16, [1, 64, 32, 24]> var_1975_cast_fp16 = reshape(shape = var_1974, x = linear_65_cast_fp16)[name = tensor<string, []>("op_1975_cast_fp16")];
            tensor<int32, [4]> value_layer_21_perm_0 = const()[name = tensor<string, []>("value_layer_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_71_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_71_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_71_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_71_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 32, 24, 64]> transpose_84 = transpose(perm = transpose_84_perm_0, x = var_1969_cast_fp16)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [1, 32, 64, 24]> transpose_83 = transpose(perm = transpose_83_perm_0, x = var_1963_cast_fp16)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [1, 32, 64, 64]> attention_scores_71_cast_fp16 = matmul(transpose_x = attention_scores_71_transpose_x_0, transpose_y = attention_scores_71_transpose_y_0, x = transpose_83, y = transpose_84)[name = tensor<string, []>("attention_scores_71_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_73_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_73_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [1, 32, 64, 64]> _inversed_attention_scores_73_cast_fp16 = mul(x = attention_scores_71_cast_fp16, y = _inversed_attention_scores_73_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_73_cast_fp16")];
            tensor<fp16, [1, 32, 64, 64]> var_1990_to_fp16 = const()[name = tensor<string, []>("op_1990_to_fp16"), val = tensor<fp16, [1, 32, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33915392)))];
            tensor<fp16, [1, 32, 64, 64]> input_219_cast_fp16 = add(x = _inversed_attention_scores_73_cast_fp16, y = var_1990_to_fp16)[name = tensor<string, []>("input_219_cast_fp16")];
            tensor<fp16, [1, 32, 64, 64]> input_221_cast_fp16 = softmax(axis = var_18, x = input_219_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")];
            tensor<bool, []> context_layer_41_transpose_x_0 = const()[name = tensor<string, []>("context_layer_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_41_transpose_y_0 = const()[name = tensor<string, []>("context_layer_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 24]> value_layer_21_cast_fp16 = transpose(perm = value_layer_21_perm_0, x = var_1975_cast_fp16)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [1, 32, 64, 24]> context_layer_41_cast_fp16 = matmul(transpose_x = context_layer_41_transpose_x_0, transpose_y = context_layer_41_transpose_y_0, x = input_221_cast_fp16, y = value_layer_21_cast_fp16)[name = tensor<string, []>("context_layer_41_cast_fp16")];
            tensor<int32, [4]> var_1995 = const()[name = tensor<string, []>("op_1995"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2000 = const()[name = tensor<string, []>("op_2000"), val = tensor<int32, [3]>([1, 64, 768])];
            tensor<fp16, [1, 64, 32, 24]> var_1996_cast_fp16 = transpose(perm = var_1995, x = context_layer_41_cast_fp16)[name = tensor<string, []>("transpose_96")];
            tensor<fp16, [1, 64, 768]> input_223_cast_fp16 = reshape(shape = var_2000, x = var_1996_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_0_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34177600)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35357312)))];
            tensor<fp16, [1, 64, 768]> linear_66_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_0_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_0_attention_output_dense_weight_to_fp16, x = input_223_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<int32, [4]> reshape_45_shape_0 = const()[name = tensor<string, []>("reshape_45_shape_0"), val = tensor<int32, [4]>([1, 1, 8, 6144])];
            tensor<fp16, [1, 1, 8, 6144]> reshape_45_cast_fp16 = reshape(shape = reshape_45_shape_0, x = linear_66_cast_fp16)[name = tensor<string, []>("reshape_45_cast_fp16")];
            tensor<int32, [4]> transpose_36_perm_0 = const()[name = tensor<string, []>("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_46_shape_0 = const()[name = tensor<string, []>("reshape_46_shape_0"), val = tensor<int32, [3]>([1, 64, 768])];
            tensor<fp16, [1, 8, 1, 6144]> transpose_36_cast_fp16 = transpose(perm = transpose_36_perm_0, x = reshape_45_cast_fp16)[name = tensor<string, []>("transpose_95")];
            tensor<fp16, [1, 64, 768]> reshape_46_cast_fp16 = reshape(shape = reshape_46_shape_0, x = transpose_36_cast_fp16)[name = tensor<string, []>("reshape_46_cast_fp16")];
            tensor<fp16, [1, 64, 768]> input_227_cast_fp16 = add(x = linear_62_cast_fp16, y = reshape_46_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")];
            tensor<int32, [1]> input_229_axes_0 = const()[name = tensor<string, []>("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_layernorm_after_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35358912)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_layernorm_after_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35360512)))];
            tensor<fp16, [1, 64, 768]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = audio_model_audio_encoder_layers_3_blocks_0_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_3_blocks_0_layernorm_after_weight_to_fp16, x = input_227_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")];
            tensor<fp16, [3072, 768]> audio_model_audio_encoder_layers_3_blocks_0_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35362112)))];
            tensor<fp16, [3072]> audio_model_audio_encoder_layers_3_blocks_0_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40080768)))];
            tensor<fp16, [1, 64, 3072]> linear_67_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_0_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_0_intermediate_dense_weight_to_fp16, x = input_229_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<string, []> input_233_mode_0 = const()[name = tensor<string, []>("input_233_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 64, 3072]> input_233_cast_fp16 = gelu(mode = input_233_mode_0, x = linear_67_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")];
            tensor<fp16, [768, 3072]> audio_model_audio_encoder_layers_3_blocks_0_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40086976)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_0_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_0_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44805632)))];
            tensor<fp16, [1, 64, 768]> linear_68_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_0_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_0_output_dense_weight_to_fp16, x = input_233_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<fp16, [1, 64, 768]> hidden_states_145_cast_fp16 = add(x = input_227_cast_fp16, y = linear_68_cast_fp16)[name = tensor<string, []>("hidden_states_145_cast_fp16")];
            tensor<int32, [1]> hidden_states_147_axes_0 = const()[name = tensor<string, []>("hidden_states_147_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_layernorm_before_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_layernorm_before_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44807232)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_layernorm_before_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_layernorm_before_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44808832)))];
            tensor<fp16, [1, 64, 768]> hidden_states_147_cast_fp16 = layer_norm(axes = hidden_states_147_axes_0, beta = audio_model_audio_encoder_layers_3_blocks_1_layernorm_before_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_3_blocks_1_layernorm_before_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("hidden_states_147_cast_fp16")];
            tensor<int32, [4]> reshape_47_shape_0 = const()[name = tensor<string, []>("reshape_47_shape_0"), val = tensor<int32, [4]>([1, 8, 1, 6144])];
            tensor<fp16, [1, 8, 1, 6144]> reshape_47_cast_fp16 = reshape(shape = reshape_47_shape_0, x = hidden_states_147_cast_fp16)[name = tensor<string, []>("reshape_47_cast_fp16")];
            tensor<int32, [4]> transpose_37_perm_0 = const()[name = tensor<string, []>("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_48_shape_0 = const()[name = tensor<string, []>("reshape_48_shape_0"), val = tensor<int32, [3]>([-1, 64, 768])];
            tensor<fp16, [1, 1, 8, 6144]> transpose_37_cast_fp16 = transpose(perm = transpose_37_perm_0, x = reshape_47_cast_fp16)[name = tensor<string, []>("transpose_94")];
            tensor<fp16, [1, 64, 768]> reshape_48_cast_fp16 = reshape(shape = reshape_48_shape_0, x = transpose_37_cast_fp16)[name = tensor<string, []>("reshape_48_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_1_attention_self_query_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_self_query_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44810432)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_attention_self_query_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_self_query_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45990144)))];
            tensor<fp16, [1, 64, 768]> linear_69_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_1_attention_self_query_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_1_attention_self_query_weight_to_fp16, x = reshape_48_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<int32, [4]> var_2110 = const()[name = tensor<string, []>("op_2110"), val = tensor<int32, [4]>([1, 64, -1, 24])];
            tensor<fp16, [1, 64, 32, 24]> var_2111_cast_fp16 = reshape(shape = var_2110, x = linear_69_cast_fp16)[name = tensor<string, []>("op_2111_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_1_attention_self_key_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_self_key_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45991744)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_attention_self_key_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_self_key_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47171456)))];
            tensor<fp16, [1, 64, 768]> linear_70_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_1_attention_self_key_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_1_attention_self_key_weight_to_fp16, x = reshape_48_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<int32, [4]> var_2116 = const()[name = tensor<string, []>("op_2116"), val = tensor<int32, [4]>([1, 64, -1, 24])];
            tensor<fp16, [1, 64, 32, 24]> var_2117_cast_fp16 = reshape(shape = var_2116, x = linear_70_cast_fp16)[name = tensor<string, []>("op_2117_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_1_attention_self_value_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_self_value_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47173056)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_attention_self_value_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_self_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48352768)))];
            tensor<fp16, [1, 64, 768]> linear_71_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_1_attention_self_value_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_1_attention_self_value_weight_to_fp16, x = reshape_48_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<int32, [4]> var_2122 = const()[name = tensor<string, []>("op_2122"), val = tensor<int32, [4]>([1, 64, -1, 24])];
            tensor<fp16, [1, 64, 32, 24]> var_2123_cast_fp16 = reshape(shape = var_2122, x = linear_71_cast_fp16)[name = tensor<string, []>("op_2123_cast_fp16")];
            tensor<int32, [4]> value_layer_perm_0 = const()[name = tensor<string, []>("value_layer_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> attention_scores_75_transpose_x_0 = const()[name = tensor<string, []>("attention_scores_75_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attention_scores_75_transpose_y_0 = const()[name = tensor<string, []>("attention_scores_75_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 32, 24, 64]> transpose_86 = transpose(perm = transpose_86_perm_0, x = var_2117_cast_fp16)[name = tensor<string, []>("transpose_91")];
            tensor<fp16, [1, 32, 64, 24]> transpose_85 = transpose(perm = transpose_85_perm_0, x = var_2111_cast_fp16)[name = tensor<string, []>("transpose_92")];
            tensor<fp16, [1, 32, 64, 64]> attention_scores_75_cast_fp16 = matmul(transpose_x = attention_scores_75_transpose_x_0, transpose_y = attention_scores_75_transpose_y_0, x = transpose_85, y = transpose_86)[name = tensor<string, []>("attention_scores_75_cast_fp16")];
            tensor<fp16, []> _inversed_attention_scores_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attention_scores_y_0_to_fp16"), val = tensor<fp16, []>(0x1.a2p-3)];
            tensor<fp16, [1, 32, 64, 64]> _inversed_attention_scores_cast_fp16 = mul(x = attention_scores_75_cast_fp16, y = _inversed_attention_scores_y_0_to_fp16)[name = tensor<string, []>("_inversed_attention_scores_cast_fp16")];
            tensor<fp16, [1, 32, 64, 64]> var_2138_to_fp16 = const()[name = tensor<string, []>("op_2138_to_fp16"), val = tensor<fp16, [1, 32, 64, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48354368)))];
            tensor<fp16, [1, 32, 64, 64]> input_239_cast_fp16 = add(x = _inversed_attention_scores_cast_fp16, y = var_2138_to_fp16)[name = tensor<string, []>("input_239_cast_fp16")];
            tensor<fp16, [1, 32, 64, 64]> input_241_cast_fp16 = softmax(axis = var_18, x = input_239_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<bool, []> context_layer_45_transpose_x_0 = const()[name = tensor<string, []>("context_layer_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> context_layer_45_transpose_y_0 = const()[name = tensor<string, []>("context_layer_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 24]> value_layer_cast_fp16 = transpose(perm = value_layer_perm_0, x = var_2123_cast_fp16)[name = tensor<string, []>("transpose_93")];
            tensor<fp16, [1, 32, 64, 24]> context_layer_45_cast_fp16 = matmul(transpose_x = context_layer_45_transpose_x_0, transpose_y = context_layer_45_transpose_y_0, x = input_241_cast_fp16, y = value_layer_cast_fp16)[name = tensor<string, []>("context_layer_45_cast_fp16")];
            tensor<int32, [4]> var_2143 = const()[name = tensor<string, []>("op_2143"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2148 = const()[name = tensor<string, []>("op_2148"), val = tensor<int32, [3]>([1, 64, 768])];
            tensor<fp16, [1, 64, 32, 24]> var_2144_cast_fp16 = transpose(perm = var_2143, x = context_layer_45_cast_fp16)[name = tensor<string, []>("transpose_90")];
            tensor<fp16, [1, 64, 768]> input_243_cast_fp16 = reshape(shape = var_2148, x = var_2144_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")];
            tensor<fp16, [768, 768]> audio_model_audio_encoder_layers_3_blocks_1_attention_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48616576)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_attention_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_attention_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49796288)))];
            tensor<fp16, [1, 64, 768]> linear_72_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_1_attention_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_1_attention_output_dense_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<int32, [4]> reshape_49_shape_0 = const()[name = tensor<string, []>("reshape_49_shape_0"), val = tensor<int32, [4]>([1, 1, 8, 6144])];
            tensor<fp16, [1, 1, 8, 6144]> reshape_49_cast_fp16 = reshape(shape = reshape_49_shape_0, x = linear_72_cast_fp16)[name = tensor<string, []>("reshape_49_cast_fp16")];
            tensor<int32, [4]> transpose_38_perm_0 = const()[name = tensor<string, []>("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> reshape_50_shape_0 = const()[name = tensor<string, []>("reshape_50_shape_0"), val = tensor<int32, [3]>([1, 64, 768])];
            tensor<fp16, [1, 8, 1, 6144]> transpose_38_cast_fp16 = transpose(perm = transpose_38_perm_0, x = reshape_49_cast_fp16)[name = tensor<string, []>("transpose_89")];
            tensor<fp16, [1, 64, 768]> reshape_50_cast_fp16 = reshape(shape = reshape_50_shape_0, x = transpose_38_cast_fp16)[name = tensor<string, []>("reshape_50_cast_fp16")];
            tensor<fp16, [1, 64, 768]> input_247_cast_fp16 = add(x = hidden_states_145_cast_fp16, y = reshape_50_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")];
            tensor<int32, [1]> input_249_axes_0 = const()[name = tensor<string, []>("input_249_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_layernorm_after_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_layernorm_after_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49797888)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_layernorm_after_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_layernorm_after_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49799488)))];
            tensor<fp16, [1, 64, 768]> input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = audio_model_audio_encoder_layers_3_blocks_1_layernorm_after_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_layers_3_blocks_1_layernorm_after_weight_to_fp16, x = input_247_cast_fp16)[name = tensor<string, []>("input_249_cast_fp16")];
            tensor<fp16, [3072, 768]> audio_model_audio_encoder_layers_3_blocks_1_intermediate_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_intermediate_dense_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49801088)))];
            tensor<fp16, [3072]> audio_model_audio_encoder_layers_3_blocks_1_intermediate_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_intermediate_dense_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54519744)))];
            tensor<fp16, [1, 64, 3072]> linear_73_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_1_intermediate_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_1_intermediate_dense_weight_to_fp16, x = input_249_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<string, []> input_253_mode_0 = const()[name = tensor<string, []>("input_253_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 64, 3072]> input_253_cast_fp16 = gelu(mode = input_253_mode_0, x = linear_73_cast_fp16)[name = tensor<string, []>("input_253_cast_fp16")];
            tensor<fp16, [768, 3072]> audio_model_audio_encoder_layers_3_blocks_1_output_dense_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_output_dense_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54525952)))];
            tensor<fp16, [768]> audio_model_audio_encoder_layers_3_blocks_1_output_dense_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_layers_3_blocks_1_output_dense_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59244608)))];
            tensor<fp16, [1, 64, 768]> linear_74_cast_fp16 = linear(bias = audio_model_audio_encoder_layers_3_blocks_1_output_dense_bias_to_fp16, weight = audio_model_audio_encoder_layers_3_blocks_1_output_dense_weight_to_fp16, x = input_253_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<fp16, [1, 64, 768]> input_257_cast_fp16 = add(x = input_247_cast_fp16, y = linear_74_cast_fp16)[name = tensor<string, []>("input_257_cast_fp16")];
            tensor<int32, [1]> last_hidden_state_1_axes_0 = const()[name = tensor<string, []>("last_hidden_state_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> audio_model_audio_encoder_norm_weight_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59246208)))];
            tensor<fp16, [768]> audio_model_audio_encoder_norm_bias_to_fp16 = const()[name = tensor<string, []>("audio_model_audio_encoder_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59247808)))];
            tensor<fp16, [1, 64, 768]> last_hidden_state_1_cast_fp16 = layer_norm(axes = last_hidden_state_1_axes_0, beta = audio_model_audio_encoder_norm_bias_to_fp16, epsilon = var_7_to_fp16, gamma = audio_model_audio_encoder_norm_weight_to_fp16, x = input_257_cast_fp16)[name = tensor<string, []>("last_hidden_state_1_cast_fp16")];
            tensor<int32, [3]> var_2199 = const()[name = tensor<string, []>("op_2199"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [5]> var_2214 = const()[name = tensor<string, []>("op_2214"), val = tensor<int32, [5]>([1, 768, 4, 2, 8])];
            tensor<fp16, [1, 768, 64]> var_2200_cast_fp16 = transpose(perm = var_2199, x = last_hidden_state_1_cast_fp16)[name = tensor<string, []>("transpose_88")];
            tensor<fp16, [1, 768, 4, 2, 8]> last_hidden_state_5_cast_fp16 = reshape(shape = var_2214, x = var_2200_cast_fp16)[name = tensor<string, []>("last_hidden_state_5_cast_fp16")];
            tensor<int32, [5]> var_2216 = const()[name = tensor<string, []>("op_2216"), val = tensor<int32, [5]>([0, 1, 3, 2, 4])];
            tensor<int32, [3]> concat_29 = const()[name = tensor<string, []>("concat_29"), val = tensor<int32, [3]>([1, 768, 64])];
            tensor<fp16, [1, 768, 2, 4, 8]> var_2217_cast_fp16 = transpose(perm = var_2216, x = last_hidden_state_5_cast_fp16)[name = tensor<string, []>("transpose_87")];
            tensor<fp16, [1, 768, 64]> input_259_cast_fp16 = reshape(shape = concat_29, x = var_2217_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")];
            tensor<int32, [1]> reduce_mean_0_axes_0 = const()[name = tensor<string, []>("reduce_mean_0_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> reduce_mean_0_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_0_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 768, 1]> reduce_mean_0_cast_fp16 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = input_259_cast_fp16)[name = tensor<string, []>("reduce_mean_0_cast_fp16")];
            tensor<int32, []> concat_30_axis_0 = const()[name = tensor<string, []>("concat_30_axis_0"), val = tensor<int32, []>(-1)];
            tensor<bool, []> concat_30_interleave_0 = const()[name = tensor<string, []>("concat_30_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 768, 1]> concat_30_cast_fp16 = concat(axis = concat_30_axis_0, interleave = concat_30_interleave_0, values = reduce_mean_0_cast_fp16)[name = tensor<string, []>("concat_30_cast_fp16")];
            tensor<int32, [2]> concat_31 = const()[name = tensor<string, []>("concat_31"), val = tensor<int32, [2]>([1, 768])];
            tensor<fp16, [1, 768]> input_261_cast_fp16 = reshape(shape = concat_31, x = concat_30_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")];
            tensor<fp16, [512, 768]> audio_projection_linear1_weight_to_fp16 = const()[name = tensor<string, []>("audio_projection_linear1_weight_to_fp16"), val = tensor<fp16, [512, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59249408)))];
            tensor<fp16, [512]> audio_projection_linear1_bias_to_fp16 = const()[name = tensor<string, []>("audio_projection_linear1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60035904)))];
            tensor<fp16, [1, 512]> linear_75_cast_fp16 = linear(bias = audio_projection_linear1_bias_to_fp16, weight = audio_projection_linear1_weight_to_fp16, x = input_261_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
            tensor<fp16, [1, 512]> input_cast_fp16 = relu(x = linear_75_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<fp16, [512, 512]> audio_projection_linear2_weight_to_fp16 = const()[name = tensor<string, []>("audio_projection_linear2_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60036992)))];
            tensor<fp16, [512]> audio_projection_linear2_bias_to_fp16 = const()[name = tensor<string, []>("audio_projection_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60561344)))];
            tensor<fp16, [1, 512]> linear_76_cast_fp16 = linear(bias = audio_projection_linear2_bias_to_fp16, weight = audio_projection_linear2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
            tensor<string, []> linear_76_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_76_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [1, 512]> audio_embeds = cast(dtype = linear_76_cast_fp16_to_fp32_dtype_0, x = linear_76_cast_fp16)[name = tensor<string, []>("cast_231")];
        } -> (audio_embeds);
}