expand fails when leading shape value is -1

Describe the bug
I am currently implementing an ONNX serializer of EBMs (explainable boosting machines). One of the needed operations is expand. The shape of the input tensors is [None, 1] and I need to expand them to [None, 3]

For example, given this input tensor:
[[0.1], [1.2], [11], [4.2]]

I expect this output tensor:

[
  [0.1, 0.1, 0.1],
  [1.2, 1.2, 1.2],
  [11, 11, 11], 
  [4.2, 4.2, 4.2]
]

When I set the shape input parameter to a tensor [-1, 3], onnx runtime fails with this error:

onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Non-zero status code returned while running Expand node. Name:'' Status Message: invalid expand shape

However, if I set it for to [4, 3], this works with the example input. Since the first dimension of my tensor is the batch size, I cannot fix the value in the input shape parameter.

Is this a bug in onnx-runtime, or should I use the operator in another way?

Urgency
None

System information

• ubuntu 18.04:
• ONNX Runtime installed from binary (pypi):
• ONNX Runtime version: 1.6.0
• Python version: 3.6.9
• Visual Studio version (if applicable):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version: None
• GPU model and memory: None

To Reproduce

sess = rt.InferenceSession('expand.onnx')
pred_onx = sess.run(None, {
            'i': [
                [0.1],
                [1.2],
                [11],
                [4.2],
            ]
        })
print(pred_onx)

Expected behavior

the result should be:

[
  [0.1, 0.1, 0.1],
  [1.2, 1.2, 1.2],
  [11, 11, 11], 
  [4.2, 4.2, 4.2]
]

Screenshots
Not applicable

Additional context
None