Fix a bug when batch type is dict and one of the values is the list

composer/core/data_spec.py

@@ -244,7 +244,16 @@ def _default_get_num_samples_in_batch(self, batch: Batch) -> int:
                                          '`get_num_samples_in_batch(your_batch) -> int` method.')
                     dim0_sizes.append(t.shape[0])
         elif isinstance(batch, dict):
-            dim0_sizes = [t.shape[0] for t in batch.values()]
+            for t in batch.values():
+                if isinstance(t, torch.Tensor):
+                    dim0_sizes.append(t.shape[0])
+                elif isinstance(t, list):
+                    dim0_sizes.append(len(t))
+                else:
+                    raise ValueError('Unable to determine the batch size as batch is a dict '
+                                     f'with an element of type {type(t)} which is not Tensor '
+                                     'or list. Please use a DataSpec and provide a '
+                                     '`get_num_samples_in_batch(your_batch) -> int` method.')
 
         if len(set(dim0_sizes)) == 1:
             return dim0_sizes[0]

tests/trainer/test_dataspec.py

@@ -86,3 +86,29 @@ def test_small_batch_at_end_warning():
 
     with pytest.warns(UserWarning, match='Cannot split tensor of length.*'):
         trainer.fit()
+
+
+@pytest.mark.parametrize(
+    'batch,num_samples',
+    [
+        [{
+            'a': torch.rand(N, 8),
+            'b': torch.rand(N, 64)
+        }, N],  # dict
+        [[{
+            'a': torch.rand(N, 8)
+        }, {
+            'c': torch.rand(N, 64)
+        }], N],  # list of dict
+        [{
+            'a': [1, 2],
+            'b': [3, 4]
+        }, 2],  # dict of lists
+        [(torch.rand(N, 8), torch.rand(N, 64)), N],  # tuple
+        [[torch.rand(N, 8), torch.rand(N, 64)], N],  # list
+        [torch.rand(N, 8), N],  # tensor
+        [torch.rand(N, 8, 4, 2), N],  # 4-dim tensor
+    ])
+def test_num_samples_in_batch(batch, num_samples):
+    data_spec = DataSpec(dataloader=DataLoader(RandomClassificationDataset(size=17), batch_size=4))
+    assert data_spec.get_num_samples_in_batch(batch) == num_samples