Cast dropout to float in RNN (#21304)
Summary:
This solves the situation where, for example, someone instantiates LSTM with `dropout=0`, a Python integer. This works fine in Python, but JIT throws a type error because it expected float but got int
Resolves https://github.com/pytorch/lockdown/issues/65
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21304
Differential Revision: D15613153
Pulled By: jamesr66a
fbshipit-source-id: eabff76e3af3de0612583b37dbc5f7eab7e248a4
