Update README.md

README.md

@@ -116,6 +116,9 @@ def check_answer(generation: str, ground_truth: str) -> bool:
     return preds == ground_truth_list
 ```
 
+Some models have specific formatting outputs for their answers, e.g. \boxed{A} or Answer: A. We recommend checking a few responses as you may notice slight variations based on this.
+This public set also has slight variations with the set used in the original paper, so while the measured capabilities are identical do not expect an exact replication of accuracy figures.
+
 
 If you'd like to use a single image model, here's a handy function to turn `image_1` and `image_2` into a single split image: