[fix] SDP syncing buffers during gradient accumulation (#1075)

- Fixes from Benjamin.

Original commit msg:
  - Fixes #1041. I just had a minute or two, hoping that it's enough :)
Co-authored-by: Min Xu <min.xu.public@gmail.com>

fairscale/nn/data_parallel/sharded_ddp.py

@@ -218,7 +218,7 @@ class ShardedDataParallel(nn.Module):
             if needs_setup:
                 self.refresh_trainable()
 
-            if self._enable_broadcast_buffers:
+            if self._enable_broadcast_buffers and not self._should_accumulate_grads:
                 # NCCL communications are on a different stream, needs to be blocking
                 # for the subsequent FW to be correct
                 self.sync_buffers(blocking=True)

tests/nn/data_parallel/test_sharded_ddp_features.py

@@ -112,6 +112,14 @@ def run_one_step(
         with ddp_model.no_sync() if grad_accumulation else suppress():
             input_tensor = torch.rand((64, 2)).to(device)
             loss = ddp_model(input_tensor).abs().sum()
+
+            # If grad_accumulation, we can check after the forward that the models are different
+            # (not synced)
+            if grad_accumulation:
+                check_same_models_across_ranks(
+                    ddp_model, dist.group.WORLD, params_should_be_equal=False, check_broadcast_buffers=True
+                )
+
             loss.backward()
         return loss