[Bug] fix a bug in post processing stage of ScienceQA. (#723)

* fix a bug in sqa_process_results

* reformat code

Author: ashun989

lmms_eval/tasks/scienceqa/utils.py

@@ -34,11 +34,11 @@ def sqa_doc_to_target(doc):
 def sqa_process_results(doc, results):
     # I know this is weird, but it's how llava parse it.
     target = sqa_doc_to_target(doc).strip().lower()
-    pred = results[0].strip().lower()
-    if pred == target:
+    pred = results[0].strip()
+    if pred.lower() == target:
         return {"exact_match": 1.0}
     # pattern: ^[A-Z]\. .*
     if len(pred) >= 2 and pred[0].isupper() and pred[1] == ".":
-        result = 1.0 if pred[0] == target else 0.0
+        result = 1.0 if pred[0].lower() == target else 0.0
         return {"exact_match": result}
     return {"exact_match": 0.0}