community: avoid double templating in langchain_prompty (#25777)

## Description

In `langchain_prompty`, messages are templated by Prompty. However, a
call to `ChatPromptTemplate` was initiating a second templating. We now
convert parsed messages to `Message` objects before calling
`ChatPromptTemplate`, signifying clearly that they are already
templated.

We also revert #25739 , which applied to this second templating, which
we now avoid, and did not fix the original issue.

## Issue

Closes #25703

libs/partners/prompty/langchain_prompty/langchain.py

@@ -1,33 +1,34 @@
-from typing import Any, Dict, Literal
+from typing import Any, Dict
 
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 from langchain_core.runnables import Runnable, RunnableLambda
 
+from .parsers import RoleMap
 from .utils import load, prepare
 
 
 def create_chat_prompt(
     path: str,
     input_name_agent_scratchpad: str = "agent_scratchpad",
-    template_format: Literal["f-string", "mustache", "jinja2"] = "f-string",
 ) -> Runnable[Dict[str, Any], ChatPromptTemplate]:
     """Create a chat prompt from a Langchain schema."""
 
     def runnable_chat_lambda(inputs: Dict[str, Any]) -> ChatPromptTemplate:
         p = load(path)
         parsed = prepare(p, inputs)
+        # Parsed messages have been templated
+        # Convert to Message objects to avoid templating attempts in ChatPromptTemplate
         lc_messages = []
         for message in parsed:
-            lc_messages.append((message["role"], message["content"]))
+            message_class = RoleMap.get_message_class(message["role"])
+            lc_messages.append(message_class(content=message["content"]))
 
         lc_messages.append(
             MessagesPlaceholder(
                 variable_name=input_name_agent_scratchpad, optional=True
             )  # type: ignore[arg-type]
         )
-        lc_p = ChatPromptTemplate.from_messages(
+        lc_p = ChatPromptTemplate.from_messages(lc_messages)
-            lc_messages, template_format=template_format
-        )
         lc_p = lc_p.partial(**p.inputs)
         return lc_p
 

libs/partners/prompty/langchain_prompty/parsers.py

@@ -1,18 +1,41 @@
 import base64
 import re
-from typing import List, Union
+from typing import Dict, List, Type, Union
 
+from langchain_core.messages import (
+    AIMessage,
+    BaseMessage,
+    FunctionMessage,
+    HumanMessage,
+    SystemMessage,
+)
 from pydantic import BaseModel
 
 from .core import Invoker, Prompty, SimpleModel
 
 
+class RoleMap:
+    _ROLE_MAP: Dict[str, Type[BaseMessage]] = {
+        "system": SystemMessage,
+        "user": HumanMessage,
+        "human": HumanMessage,
+        "assistant": AIMessage,
+        "ai": AIMessage,
+        "function": FunctionMessage,
+    }
+    ROLES = _ROLE_MAP.keys()
+
+    @classmethod
+    def get_message_class(cls, role: str) -> Type[BaseMessage]:
+        return cls._ROLE_MAP[role]
+
+
 class PromptyChatParser(Invoker):
     """Parse a chat prompt into a list of messages."""
 
     def __init__(self, prompty: Prompty) -> None:
         self.prompty = prompty
-        self.roles = ["assistant", "function", "system", "user", "human", "ai"]
+        self.roles = RoleMap.ROLES
         self.path = self.prompty.file.parent
 
     def inline_image(self, image_item: str) -> str:

libs/partners/prompty/tests/unit_tests/prompts/double_templating.prompty

@@ -0,0 +1,10 @@
+---
+name: IssuePrompt
+description: A prompt used to detect if double templating occurs
+model:
+  api: chat
+template: mustache
+---
+
+user:
+{{user_input}}

libs/partners/prompty/tests/unit_tests/test_templating.py

@@ -0,0 +1,23 @@
+from pathlib import Path
+
+import pytest
+
+from langchain_prompty import create_chat_prompt
+
+PROMPT_DIR = Path(__file__).parent / "prompts"
+
+
+def test_double_templating() -> None:
+    """
+    Assess whether double templating occurs when invoking a chat prompt.
+    If it does, an error is thrown and the test fails.
+    """
+
+    prompt_path = PROMPT_DIR / "double_templating.prompty"
+    templated_prompt = create_chat_prompt(str(prompt_path))
+    query = "What do you think of this JSON object: {'key': 7}?"
+
+    try:
+        templated_prompt.invoke(input={"user_input": query})
+    except KeyError as e:
+        pytest.fail("Double templating occurred: " + str(e))