Fix Jinja template for Advanced AI Risk (#1587)

EleutherAI · Mar 15, 2024 · dc90fec · dc90fec
1 parent 7d9922c
commit dc90fec
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/lm_eval/tasks/model_written_evals/advanced_ai_risk/_template_yaml b/lm_eval/tasks/model_written_evals/advanced_ai_risk/_template_yaml
@@ -3,7 +3,7 @@ dataset_path: EleutherAI/advanced_ai_risk
 output_type: multiple_choice
 validation_split: validation
 target_delimiter: ""
-doc_to_text: "Human: {question}\n\nAssistant:"
+doc_to_text: "Human: {{question}}\n\nAssistant:"
 doc_to_target: 0
 doc_to_choice: "{{[answer_matching_behavior, answer_not_matching_behavior]}}"
 should_decontaminate: true