diff --git a/llama_stack/providers/impls/meta_reference/datasetio/datasetio.py b/llama_stack/providers/impls/meta_reference/datasetio/datasetio.py
index 404e5c887..e3dbbc461 100644
--- a/llama_stack/providers/impls/meta_reference/datasetio/datasetio.py
+++ b/llama_stack/providers/impls/meta_reference/datasetio/datasetio.py
@@ -48,8 +48,7 @@ class PandasDataframeDataset(BaseDataset):
         self.df = None
 
     def __len__(self) -> int:
-        if self.df is None:
-            raise ValueError("Dataset not loaded. Please call .load() first")
+        assert self.df is not None, "Dataset not loaded. Please call .load() first"
         return len(self.df)
 
     def __getitem__(self, idx):
@@ -59,7 +58,7 @@ class PandasDataframeDataset(BaseDataset):
             return self.df.iloc[idx].to_dict()
 
     def load(self) -> None:
-        if self.df:
+        if self.df is not None:
             return
 
         # TODO: more robust support w/ data url
@@ -106,7 +105,6 @@ class MetaReferenceDatasetioImpl(DatasetIO, DatasetsProtocolPrivate):
         dataset_def: DatasetDef,
     ) -> None:
         dataset_impl = PandasDataframeDataset(dataset_def)
-        dataset_impl.load()
         self.dataset_infos[dataset_def.identifier] = DatasetInfo(
             dataset_def=dataset_def,
             dataset_impl=dataset_impl,
@@ -123,6 +121,8 @@ class MetaReferenceDatasetioImpl(DatasetIO, DatasetsProtocolPrivate):
         filter_condition: Optional[str] = None,
     ) -> PaginatedRowsResult:
         dataset_info = self.dataset_infos.get(dataset_id)
+        dataset_info.dataset_impl.load()
+
         if page_token is None:
             next_page_token = 0
         else: