Fix implementation of F::kl_div / F::mse_loss / F::binary_cross_entropy
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28806
Test Plan: Imported from OSS
Differential Revision: D18202859
Pulled By: yf225
fbshipit-source-id: 1aa19111cd5111dd5f2779f7f00f07f2f2e16d4d