F

src/populate.py

@@ -26,9 +26,10 @@ def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchm
     df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
     
     print(cols)
-    if version == "n_correct":
+    if version != "1_correct":
         cols.remove("VCR")
-        benchmark_cols.remove("VCR")
+        if version != "1_correct_var":
+            benchmark_cols.remove("VCR")
     else:
         for task in Detail_Tasks:
             cols.remove(task.value.col_name)