Error message on packed=True for stack exchange dataset

tests/torchtune/datasets/test_stack_exchange_paired_dataset.py

@@ -100,6 +100,16 @@ def test_dataset_get_item(self, mock_load_dataset, train_on_input):
             # Check that the input is masked
             assert sample["rejected_labels"].count(CROSS_ENTROPY_IGNORE_IDX) == 52
 
+    def test_dataset_fails_with_packed(self):
+        with pytest.raises(
+            ValueError, match="StackExchangePairedDataset does not support packing"
+        ):
+            stack_exchange_paired_dataset(
+                tokenizer=DummyTokenizer(),
+                train_on_input=True,
+                packed=True,
+            )
+
 
 class TestStackExchangePairedToMessages:
     @pytest.fixture

torchtune/datasets/_stack_exchange_paired.py

@@ -78,6 +78,7 @@ def stack_exchange_paired_dataset(
     source: str = "lvwerra/stack-exchange-paired",
     column_map: Optional[Dict[str, str]] = None,
     train_on_input: bool = False,
+    packed: bool = False,
     filter_fn: Optional[Callable] = None,
     split: str = "train",
     **load_dataset_kwargs: Dict[str, Any],
@@ -101,17 +102,25 @@ def stack_exchange_paired_dataset(
             Keys should be "prompt", "chosen", and "rejected" and values should be the actual column names.
             Default is None, keeping the default column names.
         train_on_input (bool): Whether the model is trained on the prompt or not. Default is False.
+        packed (bool): Whether or not to pack the dataset to ``max_seq_len`` prior to training. Default is False and this
+            is currently not supported for this dataset. A ValueError will be raised if this is set to True.
         filter_fn (Optional[Callable]): callable used to filter the dataset prior to any pre-processing. See
             the Hugging Face `docs <https://huggingface.co/docs/datasets/v2.20.0/process#select-and-filter>`_ for more
             details.
         split (str): ``split`` argument for ``datasets.load_dataset``. You can use this argument to load a subset
             of a given split, e.g. ``split="train[:10%]"``. Default is "train".
         **load_dataset_kwargs (Dict[str, Any]): additional keyword arguments to pass to ``load_dataset``.
 
+    Raises:
+        ValueError: If ``packed`` is True, they are not supported for preference datasets yet.
+
     Returns:
         PreferenceDataset: The preference dataset built from source paired data.
     """
 
+    if packed:
+        raise ValueError("StackExchangePairedDataset does not support packing.")
+
     column_map = column_map or {
         "prompt": "question",
         "chosen": "response_j",