Set shuffle=True to test again #6531

microsoft · Nov 4, 2024 · 29587e2 · 29587e2
1 parent 9068acb
commit 29587e2
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/deepspeed/runtime/engine.py b/deepspeed/runtime/engine.py
@@ -1774,7 +1774,7 @@ def deepspeed_io(self,
                 dataset,
                 num_replicas=data_parallel_world_size,
                 rank=data_parallel_rank,
-                shuffle=False,
+                shuffle=True,
             )
 
         deepspeed_dataloader_config = {}

diff --git a/deepspeed/runtime/pipe/engine.py b/deepspeed/runtime/pipe/engine.py
@@ -266,7 +266,7 @@ def _build_data_iter(self, dataset):
         sampler = torch.utils.data.distributed.DistributedSampler(dataset,
                                                                   num_replicas=self.dp_world_size,
                                                                   rank=self.mpu.get_data_parallel_rank(),
-                                                                  shuffle=False)
+                                                                  shuffle=True)
         # Build a loader and make it repeating.
         pipe_dataloader = self.deepspeed_io(dataset, data_sampler=sampler)
         pipe_dataloader = RepeatingLoader(pipe_dataloader)