Fix --ddp-backend=no_c10d for params that don't require grads

Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/388

Reviewed By: theweiho

Differential Revision: D13244869

fbshipit-source-id: d22c18f63f9a691ccc7245e06bc9a5b776a192b5

fairseq/legacy_distributed_data_parallel.py

@@ -166,4 +166,5 @@ class LegacyDistributedDataParallel(nn.Module):
             def allreduce_hook(*unused):
                 Variable._execution_engine.queue_callback(reduction_fn)
 
-            p.register_hook(allreduce_hook)
+            if p.requires_grad:
+                p.register_hook(allreduce_hook)