How about this wording?

    If alpha is smaller than 0.1, an occasional OverflowError can
    arise when the variate exceeds the range of Python float.

I like using 0.1 because it's easy and gives us some wiggle room.  The actual cutoff is lower:

    >>> alpha = 0.05079
    >>> sys.float_info.epsilon ** (-1.0 / alpha)

    >>> alpha = 0.05078
    >>> sys.float_info.epsilon ** (-1.0 / alpha)
    Traceback (most recent call last):
       File "<pyshell#48>", line 1, in <module>
         sys.float_info.epsilon ** (-1.0 / alpha)
    1.590779741838475e+308