diff --git a/src/transformers/pipelines/base.py b/src/transformers/pipelines/base.py
index 53baca01342687..11eba10589ef3c 100644
--- a/src/transformers/pipelines/base.py
+++ b/src/transformers/pipelines/base.py
@@ -70,18 +70,30 @@ def _pad(items, key, padding_value, padding_side):
     if isinstance(items[0][key], torch.Tensor):
         # Others include `attention_mask` etc...
         shape = items[0][key].shape
-        if len(shape) != 2:
+        dim = len(shape)
+        if dim == 4:
             # This is probable image so padding shouldn't be necessary
+            # B, C, H, W
             return torch.cat([item[key] for item in items], dim=0)
-        max_length = max(item[key].shape[-1] for item in items)
+        max_length = max(item[key].shape[1] for item in items)
         dtype = items[0][key].dtype
-        tensor = torch.zeros((batch_size, max_length), dtype=dtype) + padding_value
+
+        if dim == 2:
+            tensor = torch.zeros((batch_size, max_length), dtype=dtype) + padding_value
+        elif dim == 3:
+            tensor = torch.zeros((batch_size, max_length, shape[-1]), dtype=dtype) + padding_value
 
         for i, item in enumerate(items):
-            if padding_side == "left":
-                tensor[i, -len(item[key][0]) :] = item[key][0].clone()
-            else:
-                tensor[i, : len(item[key][0])] = item[key][0].clone()
+            if dim == 2:
+                if padding_side == "left":
+                    tensor[i, -len(item[key][0]) :] = item[key][0].clone()
+                else:
+                    tensor[i, : len(item[key][0])] = item[key][0].clone()
+            elif dim == 3:
+                if padding_side == "left":
+                    tensor[i, -len(item[key][0]) :, :] = item[key][0].clone()
+                else:
+                    tensor[i, : len(item[key][0]), :] = item[key][0].clone()
         return tensor
     else:
         return [item[key] for item in items]
@@ -1022,7 +1034,7 @@ def get_iterator(
         final_iterator = PipelineIterator(model_iterator, self.postprocess, postprocess_params)
         return final_iterator
 
-    def __call__(self, inputs, *args, num_workers=2, batch_size=1, **kwargs):
+    def __call__(self, inputs, *args, num_workers=0, batch_size=1, **kwargs):
         if args:
             logger.warning(f"Ignoring args : {args}")
         preprocess_params, forward_params, postprocess_params = self._sanitize_parameters(**kwargs)
diff --git a/tests/test_pipelines_common.py b/tests/test_pipelines_common.py
index 13ca39c63dc3ca..e64d4b8c09f9fb 100644
--- a/tests/test_pipelines_common.py
+++ b/tests/test_pipelines_common.py
@@ -307,14 +307,23 @@ def test_pipeline_image_padding(self):
                 torch.zeros((2, 3, 10, 10)),
             )
         )
+
+    @require_torch
+    def test_pipeline_offset_mapping(self):
+        import torch
+
+        items = [
+            {
+                "offset_mappings": torch.zeros([1, 11, 2], dtype=torch.long),
+            },
+            {
+                "offset_mappings": torch.zeros([1, 4, 2], dtype=torch.long),
+            },
+        ]
+
         self.assertTrue(
             torch.allclose(
-                _pad(items, "input_ids", 10, "left"),
-                torch.LongTensor([[10, 10, 1, 23, 24, 2], [1, 23, 24, 43, 44, 2]]),
-            )
-        )
-        self.assertTrue(
-            torch.allclose(
-                _pad(items, "attention_mask", 0, "right"), torch.LongTensor([[0, 1, 1, 0, 0, 0], [0, 1, 1, 1, 1, 0]])
-            )
+                _pad(items, "offset_mappings", 0, "right"),
+                torch.zeros((2, 11, 2), dtype=torch.long),
+            ),
         )