make --chat and --use-grammar mutually exclusive for evaluation task

task.py

@@ -113,6 +113,7 @@ class SentimentAnalysis(Task):
         response = self.model(
             prompt=CLASSIFICATION_PROMPT.format(instruction=prompt, input=text),
             grammar=sa_grammar_fn() if self.use_grammar else None,
+            chat=False if self.use_grammar else True,
         )
 
         if not self.use_grammar:
@@ -259,6 +260,7 @@ class QuestionAnswering(Task):
                 question=question,
             ),
             grammar=grammar,
+            chat=False if self.use_grammar else True,
         )
 
         if not self.use_grammar: