set the default attn to flash attention

aria/model/configuration_aria.py

@@ -73,7 +73,11 @@ def __init__(
         self.image_token_index = image_token_index
 
         attn_implementation = kwargs.pop("attn_implementation", None)
-        self._attn_implementation = attn_implementation
+
+        # Set the default attention implementation to flash_attention_2 if not specified
+        self._attn_implementation = (
+            "flash_attention_2" if attn_implementation is None else attn_implementation
+        )
 
         # Convert the keys and values of projector_patch_to_query_dict to integers
         # This ensures consistency even if they were provided as strings