Types troubles

[GurovNik]

Here is a place for real troubles.

At 1 we set the type first time. At 2 - the second time. At 3 there is the craziest thing. For different types, we should define variables of different names.
Example:
float_val = val
int_val = val
The function should have one variable name for all cases. That is the fundamental thing for comfortable usage of the framework.
More exciting is that:
int_val = val
int64_val = val
are different things.
More specifically:
tensor = hs.TensorProto(dtype=hs.DT_FLOAT, tensor_shape=tensor_shape, int_val=val.flatten())
tensor = hs.TensorProto(dtype=hs.DT_FLOAT, tensor_shape=tensor_shape, int64_val=val.flatten())
One of the lines is valid and second is not depending on the situation. End there is no hints that the problem is here.
The main issue - having all these things in the mind during DS developing is very painful. Especially when logs have no hints on the cause of the problem. Of course, there is documentation. But from my point of view service should be intuitive. Looking through documentation each time user deploy a model is a bad user exp.