c# onnxruntime bug

[williamlzw]

c# onnxruntime 1.21.0
model url:https://hf-mirror.com/H5N1AIDS/F5-TTS-ONNX/blob/main/F5_TTS_CUDA_F16.zip
https://hf-mirror.com/H5N1AIDS/F5-TTS-ONNX/blob/main/F5-TTS-ONNX-Inference.py
line 213:

noise, rope_cos, rope_sin, cat_mel_text, cat_mel_text_drop, ref_signal_len = ort_session_A.run(
 [out_name_A0, out_name_A1, out_name_A2, out_name_A3, out_name_A4, out_name_A5],
   {
   in_name_A0: audio,
   in_name_A1: text_ids,
   in_name_A2: max_duration
   })

print(audio.shape, text_ids.shape, max_duration.shape)
print(audio.dtype, text_ids.dtype, max_duration.dtype)
(1, 1, 288000) (1, 343) ()
int16 int32 int64

var audio = OrtValue.CreateTensorValueFromMemory(
    waveform.data<Int16>().ToArray(), new long[] { 1, 1, waveform.shape[0] });
var textIds = OrtValue.CreateTensorValueFromMemory(
    textIdsTensor.data<int>().ToArray(), new long[] { 1, textIdsTensor.shape[1] });
var maxDuration = OrtValue.CreateTensorValueFromMemory(
    new long[]{ maxDurationValue }, new long[] { 1 });
Console.WriteLine(string.Join(",", audio.GetTensorTypeAndShape().Shape.Select(x => x.ToString())));
Console.WriteLine(string.Join(",", textIds.GetTensorTypeAndShape().Shape.Select(x => x.ToString())));
Console.WriteLine(string.Join(",", maxDuration.GetTensorTypeAndShape().Shape.Select(x => x.ToString())));
var inputsValues = new List<OrtValue>
 {
     audio ,
     textIds ,
     maxDuration 
 };

var inputsNames = new List<string>
 {
     "audio",
      "text_ids",
      "max_duration"
 };
var results = _session_a.Run(new RunOptions(), inputsNames, inputsValues, _session_a.OutputNames);
var noise = results[0].GetTensorDataAsSpan<float>().ToArray();

1,1,288000
1,335
1

[yuslepukhin]

What makes you think it is a bug, and not the input/model problem?

[williamlzw]

The same ONNX model can be correctly inferred in the Python version, but encounters an error in the C# version.

[williamlzw]

Microsoft.ML.OnnxRuntime.OnnxRuntimeException:“[ErrorCode:RuntimeException] Non-zero status code returned while running Concat node. Name:'/Concat_4' Status Message: D:\a_work\1\s\onnxruntime\core\providers\cpu\tensor\concat.cc:141 onnxruntime::ConcatBase::PrepareForCompute input_rank == reference_rank was false. Ranks of input data are different, cannot concatenate them. expected rank: 1 got: 2
”

[williamlzw]

I use this new file to convert model to onnx, it can be run on C#.
https://github.com/DakeQQ/F5-TTS-ONNX/blob/main/Export_ONNX/F5_TTS/Export_F5.py