corrected tensor calculation logic

deeplake/enterprise/test_pytorch.py

@@ -6,7 +6,11 @@
 
 from deeplake.util.remove_cache import get_base_storage
 from deeplake.core.index.index import IndexEntry
-from deeplake.tests.common import requires_torch, requires_libdeeplake
+from deeplake.tests.common import (
+    requires_torch,
+    requires_libdeeplake,
+    convert_data_according_to_torch_version,
+)
 from deeplake.core.dataset import Dataset
 from deeplake.constants import KB
 
@@ -458,7 +462,7 @@ def test_pytorch_decode(hub_cloud_ds, compressed_image_paths, compression):
     ptds = hub_cloud_ds.dataloader().pytorch(decode_method={"image": "tobytes"})
 
     for i, batch in enumerate(ptds):
-        image = batch["image"]
+        image = convert_data_according_to_torch_version(batch["image"])
         assert isinstance(image, bytes)
         if i < 5 and not compression:
             np.testing.assert_array_equal(

deeplake/integrations/pytorch/shuffle_buffer.py

@@ -118,6 +118,10 @@ def _num_torch_tensors(self, sample):
             return 0
         if isinstance(sample, TorchTensor):
             return 1
+        elif isinstance(sample, bytes):
+            return 0
+        elif isinstance(sample, str):
+            return 0
         elif isinstance(sample, dict):
             return sum(self._num_torch_tensors(tensor) for tensor in sample.values())
         elif isinstance(sample, Sequence):

deeplake/requirements/tests.txt

@@ -16,3 +16,5 @@ boto3-stubs[essential]
 lz4
 rich
 wandb
+
+pandas; python_version >= '3.11' and sys_platform == 'win32'

deeplake/tests/common.py

@@ -142,9 +142,7 @@ def __exit__(self, *args, **kwargs):
 
 
 def convert_data_according_to_torch_version(batch):
-    import torch
-
-    if torch.__version__ < "2.0.0":
+    if isinstance(batch, List):
         return batch[0]
     else:
         return batch