Comment out data loader bug fix

metaseq/data/iterators.py

@@ -358,23 +358,24 @@ def load_state_dict(self, state_dict):
             self._itr = self._get_iterator_for_epoch(self.epoch)
             self._itr.n = n
 
-            if True:
-                # Epilogue bug fixup
-                # Warning: this fix is not correct for the last ~1% of an epoch, but it only needs to be
-                # applied once earlier in the epoch to fix any data loaders with incorrect data.
-                num_workers = self._itr.num_workers
-                batch_size = self._itr.batch_size * self._itr.num_shards
-                if sum(sequences_consumed) != n * batch_size:
-                    logger.warning(
-                        f"{distributed_utils.get_global_rank()}: Sequences appear corrupted: "
-                        f"{n}*{batch_size} != sum({sequences_consumed})"
-                    )
-                    each, left = divmod(n, num_workers)
-                    sequences_consumed = [
-                        batch_size * (each + (1 if i < left else 0))
-                        for i in range(num_workers)
-                    ]
-                assert sum(sequences_consumed) == n * batch_size
+            # See https://github.com/facebookresearch/metaseq/pull/566/files for context.
+            # if True:
+            #     # Epilogue bug fixup
+            #     # Warning: this fix is not correct for the last ~1% of an epoch, but it only needs to be
+            #     # applied once earlier in the epoch to fix any data loaders with incorrect data.
+            #     num_workers = self._itr.num_workers
+            #     batch_size = self._itr.batch_size * self._itr.num_shards
+            #     if sum(sequences_consumed) != n * batch_size:
+            #         logger.warning(
+            #             f"{distributed_utils.get_global_rank()}: Sequences appear corrupted: "
+            #             f"{n}*{batch_size} != sum({sequences_consumed})"
+            #         )
+            #         each, left = divmod(n, num_workers)
+            #         sequences_consumed = [
+            #             batch_size * (each + (1 if i < left else 0))
+            #             for i in range(num_workers)
+            #         ]
+            #     assert sum(sequences_consumed) == n * batch_size
 
             self._itr.sequences_consumed = sequences_consumed
             self._itr.next_worker = next_worker