diff --git a/evoprompt/task/task.py b/evoprompt/task/task.py
index cfc79a208161243ce89ecbf9ad97d95d11091d6a..860a620e7be456e4e611001a9e39e265b3327d20 100644
--- a/evoprompt/task/task.py
+++ b/evoprompt/task/task.py
@@ -328,7 +328,7 @@ class Task(metaclass=ABCMeta):
             lambda _, idx: idx in samples_ids, with_indices=True
         )
         # remove demonstration samples from validation set
-        remaining_dataset = self.dataset.filter(
+        remaining_dataset = dataset.filter(
             lambda _, idx: idx not in samples_ids, with_indices=True
         )
         return demonstration_samples, remaining_dataset