update llm command generator

rasa/dialogue_understanding/generator/llm_command_generator.py

@@ -38,6 +38,7 @@
 from rasa.shared.nlu.training_data.training_data import TrainingData
 from rasa.shared.utils.llm import (
     DEFAULT_OPENAI_CHAT_MODEL_NAME_ADVANCED,
+    get_prompt_template,
     llm_factory,
     tracker_as_readable_transcript,
     sanitize_message_for_prompt,
@@ -82,7 +83,10 @@ def __init__(
         resource: Resource,
     ) -> None:
         self.config = {**self.get_default_config(), **config}
-        self.prompt_template = self.config["prompt"]
+        self.prompt_template = get_prompt_template(
+            DEFAULT_COMMAND_PROMPT_TEMPLATE,
+            self.config.get("prompt"),
+        )
         self._model_storage = model_storage
         self._resource = resource
 

rasa/shared/utils/llm.py

@@ -10,6 +10,7 @@
 from rasa.shared.core.trackers import DialogueStateTracker
 from rasa.shared.core.events import BotUttered, UserUttered
 from rasa.shared.engine.caching import get_local_cache_location
+import rasa.shared.utils.io
 
 
 structlogger = structlog.get_logger()
@@ -208,3 +209,22 @@ def embedder_factory(
         return embeddings_cls(**parameters)
     else:
         raise ValueError(f"Unsupported embeddings type '{typ}'")
+
+
+def get_prompt_template(
+    default_prompt_template: Text, jinja_file_path: Optional[Text]
+) -> Text:
+    """Returns the prompt template.
+
+    Args:
+        default_prompt_template: the default prompt template
+        jinja_file_path: the path to the jinja file
+
+    Returns:
+        The prompt template.
+    """
+    return (
+        rasa.shared.utils.io.read_file(jinja_file_path)
+        if jinja_file_path is not None
+        else default_prompt_template
+    )