change prompt formatting

lm_eval/tasks/nq_open/nq_open.yaml

@@ -3,9 +3,10 @@ dataset_path: nq_open
 output_type: greedy_until
 training_split: train
 validation_split: validation
-description: "Answer these questions:\n\n"
+description: "Answer these questions:\n"
 doc_to_text: "Q: {{question}}?\nA:"
 doc_to_target: "{{answer}}" # TODO: should be multi-target
+fewshot_delimiter: "\n"
 generation_kwargs:
   until:
     - "\n"