Inference result is different between keras model and c++

[chojuahn]

Using the sample code with a single Dense layer, the result is different between keras model and c++.
keras model: -0.03101047
c++: -0.0310103893

I would like to know where this decimal difference comes from.
I'm attaching .h5 and model file for easier reproducing.

keras: 2.3.0
tensorflow-gpu: 2.1.0
example.zip