EyeLike with dynamic shape results in error

[dashesy]

Describe the bug
I needed to export eye so used this:

    def eye(g, n, m, dtype=None, layout=None, device=None, pin_memory=False):
        s0 = g.op("Unsqueeze", n, axes_i=[0])
        shape = g.op("Concat", s0, s0, axis_i=0)
        tensor = zeros(g, shape, dtype, layout, device)
        return g.op("EyeLike", tensor)

Now I get this when I run the exported model

RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running EyeLike node. Name:'__EyeLike_1523' Status Message: /onnxruntime_src/onnxruntime/core/framework/op_kernel.cc:44 OrtValue* onnxruntime::OpKernelContext::OutputMLValue(int, const onnxruntime::TensorShape&, size_t) status.IsOK() was false. Shape mismatch attempting to re-use buffer. {} != {83,83}. Validate usage of dim_value (values should be > 0) and dim_param (all values with the same string should equate to the same size) in shapes in the model.

Before I got that as warnings, this time it is an error

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 16.04
• ONNX Runtime installed from (source or binary): 1.4
• ONNX Runtime version: 1.4
• Python version: 3.6
• Visual Studio version (if applicable):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:

To Reproduce

• Describe steps/code to reproduce the behavior.
• Attach the ONNX model to the issue (where applicable) to expedite investigation.

Expected behavior
No error

Screenshot

[askhade]

@dashesy : Can you share an example model for this?

[stale]

This issue has been automatically marked as stale due to inactivity and will be closed in 7 days if no further activity occurs. If further support is needed, please provide an update and/or more details.