Fixing mask multiplication with grad tensors

Grads can be None type. Adding this fix to skip multiplication with masks if this is the case.

apex/contrib/sparsity/asp.py

@@ -131,6 +131,7 @@ class ASP:
             # prune gradients before step method
             with torch.no_grad():
                 for module_name, module, p_name, p, mask, pruned in cls.__sparse_parameters:
+                    if p.grad is not None: #thx pjudd
                         p.grad.mul_(mask)
             # call original optimizer step method
             rval = opt_self.__step(*args, **kwargs)