reward model doesn't work well with batched

src/axolotl/prompt_strategies/bradley_terry/chat_template.py

@@ -21,6 +21,10 @@ class BTChatTemplateStrategy(ChatTemplateStrategy):
     Bradley-Terry reward model pairwise chat template prompt strategy.
     """
 
+    @property
+    def supports_batched(self) -> bool:
+        return False
+
     def _tokenize_single_prompt(self, prompt):
         """
 

src/axolotl/prompt_strategies/chat_template.py

@@ -239,7 +239,7 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
         Public method that can handle either a single prompt or a batch of prompts.
         """
 
-        if not self.is_prompt_batched(prompt):
+        if not self.is_prompt_batched(prompt) or not self.supports_batched:
             return self._tokenize_single_prompt(prompt)
 
         res = defaultdict(lambda: [])