This pull request fixes a bug / typo in layers/Embed.py on lines 13 and 50.

Torch tensors do not have an attribute called require_grad, and therefore pe.require_grad = False and w.require_grad = False calls were resulting in the insertion of an additional require_grad attribute to the instantiated tensor, that did nothing.

The correct attribute to track gradients is Torch is called requires_grad, and in this PR both lines (13 and 50) are changed to reflect this.

The existing bug / typo is not expected to have affected any code behavior, as when a tensor is created using torch.zeros, it by default comes with requires_grad attribute set to False anyways. But for the sake of clarity and preventing any future confusion it is worthwhile to fix the bug anyways.