Half of the length that correct output shape

[ElinLiu0]

Describe the issue

I'm trying using Numpy.NET and ONNXRuntime C# to inference model.

It's actually runs on .NET with result shape(1,3590,768) but in Python is (1,7180,768).

To reproduce

Here is code example for inference:

Tensor<float> audioTensor = new DenseTensor<float>(audio16k.GetData<float>(), new int[] { 1, audio16k.len });
var inputs = new List<NamedOnnxValue>
{
    NamedOnnxValue.CreateFromTensor("feats", audioTensor)
};
using IDisposableReadOnlyCollection<DisposableNamedOnnxValue> hubertOutput = this.HubertModle?.Run(inputs);
NDarray hubertOutputArray = np.array(hubertOutput[0].AsTensor<float>().ToArray(), dtype: np.float32);
hubertOutputArray = hubertOutputArray.reshape(1, hubertOutputArray.shape[0] / 768 , 768).transpose(0,2,1);
hubertOutputArray = np.repeat(hubertOutputArray, [2], axis: 2).transpose(0,2,1).astype(np.float32);
int hubertLength = hubertOutputArray.shape[1];

Urgency

No response

Platform

Windows

OS Version

Windows11 24H2

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.16.0

ONNX Runtime API

C#

Architecture

X64

Execution Provider

CUDA

Execution Provider Library Version

CUDA 11.6

[yuslepukhin]

Try this: https://onnxruntime.ai/docs/tutorials/csharp/basic_csharp.html

[ElinLiu0]

Try this: https://onnxruntime.ai/docs/tutorials/csharp/basic_csharp.html

Thanks,I'll try this and bring your reply ASAP.

[skottmckay]

The output shape and data type is defined by the model. View the ONNX model in Netron to check those details.

[ElinLiu0]

The output shape and data type is defined by the model. View the ONNX model in Netron to check those details.

I definitely know the output shape.But it's weried that i got differenct output shape between two language sdks.

I'll try @yuslepukhin's tutorial later than check if i'm using a wrong way to get tensor result.