Pydantic inconsistent and automatic conversion between float and int

I am using pydantic python package in FastAPI for a webapp, and I noticed there are some inconsistent float-int conversion with different typing checks. For example:

class model(BaseModel):
    data: Optional[Union[int, float]] = None
m = model(data=3.33)
m.data --> 3.33

class model(BaseModel):
    data: Optional[Union[int, float, str]] = None
m = model(data=3.33)
m.data --> 3

class model(BaseModel):
    data: Union[int, float, str] = None
m = model(data=3.33)
m.data --> 3

class model(BaseModel):
    data: Union[str, int, float] = None
m = model(data=3.33)
m.data --> '3.33'

As shown here, different orders/combinations of typings have different behaviors.

I checked out thread https://github.com/samuelcolvin/pydantic/issues/360, and https://github.com/samuelcolvin/pydantic/issues/284, but they seem not to be the exact same problem.

What causes such behavior under the hood? Is there a specific reason for this? Or did I do anything wrong/inappropriate here?

I'm using python 3.8, pydantic 1.8.2

Thank you for helping!