Why specify additional details when using KerasLayer?

[balaji.ambresh]

The last exercise asks to specify input_shape, output_shape and dtype parameters. We can use either the model loaded from the disk or use the handle. Could someone please shed light on why these are required?

[gopiramena]

These are optional, but considered best and safe to pass. to avoid loading wrong model.
From the docs,
“Note: to work-around missing shape inference functionalities from functions created from FunctionDefs, in rare cases one has to pass an ‘output_shape’ and potentially ‘input_shape’ and ‘dtype’”