fix batchsampler does not work correctly (#20327)

* fix batchsampler does not work correctly

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* add batch sampler shuffle state test

Author: dadwadw233

src/lightning/pytorch/utilities/data.py

@@ -349,7 +349,8 @@ def _is_dataloader_shuffled(dataloader: object) -> bool:
     if not hasattr(dataloader, "sampler"):
         # shuffling is enabled via a sampler. No sampler, no shuffling
         return False
-    sampler = dataloader.sampler
+    batch_sampler = dataloader.batch_sampler
+    sampler = batch_sampler.sampler if batch_sampler is not None else dataloader.sampler
     if isinstance(sampler, SequentialSampler):
         return False
     return isinstance(sampler, RandomSampler)

tests/tests_pytorch/utilities/test_data.py

@@ -12,6 +12,7 @@
 from lightning.pytorch.trainer.states import RunningStage
 from lightning.pytorch.utilities.data import (
     _get_dataloader_init_args_and_kwargs,
+    _is_dataloader_shuffled,
     _update_dataloader,
     extract_batch_size,
     has_len_all_ranks,
@@ -20,7 +21,7 @@
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
 from lightning_utilities.test.warning import no_warning_call
 from torch import Tensor
-from torch.utils.data import BatchSampler, DataLoader, RandomSampler
+from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
 
 
 def test_extract_batch_size():
@@ -304,6 +305,31 @@ def __init__(self, extra_arg):
         _ = _update_dataloader(dataloader, dataloader.sampler, mode=RunningStage.PREDICTING)
 
 
+def test_batch_sampler_shuffle_setting():
+    """Test whether the `shuffle` state is correctly set in the `BatchSampler`."""
+
+    random_sampler = RandomSampler(range(10))
+    seq_sampler = SequentialSampler(range(10))
+    shuffled_dataloader = DataLoader(
+        range(10), batch_sampler=BatchSampler(random_sampler, batch_size=2, drop_last=False)
+    )
+    sequential_dataloader = DataLoader(
+        range(10), batch_sampler=BatchSampler(seq_sampler, batch_size=2, drop_last=False)
+    )
+
+    # if batch_size is 1, the pytorch init a default SequentialSampler and set BatchSampler to None
+    single_dataloader = DataLoader(range(10), batch_sampler=BatchSampler(seq_sampler, batch_size=1, drop_last=False))
+    assert _is_dataloader_shuffled(shuffled_dataloader)
+    assert not _is_dataloader_shuffled(sequential_dataloader)
+    assert not _is_dataloader_shuffled(single_dataloader)
+
+    # if batch_size is 1, and no batch_sampler is set, the pytorch will set BatchSampler to None
+    single_dataloader = DataLoader(range(10), batch_size=1)
+    shuffled_single_dataloader = DataLoader(range(10), batch_size=1, shuffle=True)
+    assert not _is_dataloader_shuffled(single_dataloader)
+    assert _is_dataloader_shuffled(shuffled_single_dataloader)
+
+
 @pytest.mark.parametrize("mode", [RunningStage.TRAINING, RunningStage.PREDICTING, RunningStage.TESTING])
 def test_dataloader_kwargs_replacement_with_iterable_dataset(mode):
     """Test that DataLoader kwargs are not replaced when using Iterable Dataset."""