diff --git a/community_tasks/swiss_legal_evals.py b/community_tasks/swiss_legal_evals.py index ccc18def..6f7c8ab5 100644 --- a/community_tasks/swiss_legal_evals.py +++ b/community_tasks/swiss_legal_evals.py @@ -268,7 +268,7 @@ def compute( **kwargs, ) -> dict[str, float]: logger.info(f"Judging {len(formatted_docs)} samples with {self.short_judge_name}...") - questions = [formatted_doc.specific["question"] for formatted_doc in formatted_docs] + questions = [formatted_doc.specific["source"] for formatted_doc in formatted_docs] options = [formatted_doc.choices for formatted_doc in formatted_docs] golds = [formatted_doc.get_golds()[0] for formatted_doc in formatted_docs] predictions = [response[0].result for response in responses]