Refactor analysis for per turn eval

parlai/crowdsourcing/tasks/model_chat/model_chat_blueprint.py

@@ -219,6 +219,7 @@ def __init__(
         left_pane_path = os.path.expanduser(args.blueprint.left_pane_text_path)
         with open(left_pane_path, "r") as left_pane_file:
             self.left_pane_text = left_pane_file.read()
+        self.format_left_pane_text()
         self.annotations_config: Optional[str] = None
         if args.blueprint.get("annotations_config_path", "") != "":
             annotations_config_path = os.path.expanduser(
@@ -253,6 +254,12 @@ def __init__(
             }
         )
 
+    def format_left_pane_text(self):
+        """
+        Modifies self.left_pane_text for code injection
+        """
+        pass
+
     @abstractmethod
     def _get_shared_models(self, args: "DictConfig") -> Dict[str, dict]:
         """

parlai/crowdsourcing/tasks/model_chat/task_config/model_opts.yaml

@@ -1,2 +1,2 @@
 blender_90M: >
-    --model-file zoo:blender/blender_90M/model
+    --model-file zoo:blender/blender_90M/model