x

libs/langchain/langchain/automaton/chat_agent.py

@@ -3,22 +3,23 @@ from __future__ import annotations
 
 import ast
 import re
-from typing import Sequence, Union, Optional
+from typing import Sequence, Union, Optional, List
 
 from langchain.automaton.prompt_generators import MessageLogPromptValue
 from langchain.automaton.runnables import create_llm_program
 from langchain.automaton.typedefs import (
     MessageLog,
-    AgentFinish, MessageLike, FunctionCall,
+    AgentFinish,
+    MessageLike,
+    FunctionCall,
+    FunctionResult,
 )
 from langchain.schema.language_model import BaseLanguageModel
-from langchain.schema.messages import BaseMessage
+from langchain.schema.messages import SystemMessage, BaseMessage
 from langchain.tools import BaseTool
 
 
-class ActionParser:
-    """A utility class to encode and decode action blocks."""
-
+class ActionEncoder:
     def __init__(self) -> None:
         """Initialize the ActionParser."""
         self.pattern = re.compile(r"<action>(?P<action_blob>.*?)<\/action>", re.DOTALL)
@@ -40,6 +41,29 @@ class ActionParser:
         else:
             return None
 
+    def encode_as_str(self, function_call: FunctionCall) -> str:
+        """Encode the action."""
+        if function_call.name == "Final Answer":
+            return f"<action>{{'action': 'Final Answer', 'action_input': '{function_call.arguments}'}}</action>"
+        return f"<action>{{'action': '{function_call.name}', 'action_input': {function_call.arguments}}}</action>"
+
+
+def prompt_generator(log: MessageLog) -> List[BaseMessage]:
+    """Generate a prompt from a log of message like objects."""
+    messages = []
+    for message in log.messages:
+        if isinstance(message, BaseMessage):
+            messages.append(message)
+        elif isinstance(message, FunctionResult):
+            messages.append(
+                SystemMessage(
+                    content=f"Observation: {message.result}",
+                )
+            )
+        else:
+            pass
+    return messages
+
 
 class ChatAgent:
     """An agent for chat models."""
@@ -52,11 +76,12 @@ class ChatAgent:
         max_iterations: int = 10,
     ) -> None:
         """Initialize the chat automaton."""
+        action_encoder = ActionEncoder()
         self.llm_program = create_llm_program(
             llm,
             prompt_generator=MessageLogPromptValue.from_message_log,
             tools=tools,
-            parser=OpenAIFunctionsParser(),
+            parser=action_encoder.decode,
         )
         self.max_iterations = max_iterations
 

libs/langchain/langchain/automaton/openai_agent.py

@@ -3,10 +3,8 @@ from __future__ import annotations
 import json
 from typing import Sequence, List
 
-from langchain.automaton.prompt_generators import MessageLogPromptValue
 from langchain.automaton.runnables import create_llm_program
 from langchain.automaton.typedefs import (
-    MessageLike,
     MessageLog,
     AgentFinish,
     FunctionCall,
@@ -43,13 +41,19 @@ class OpenAIFunctionsParser(BaseGenerationOutputParser):
         )
 
 
-def message_adapter(message: MessageLike) -> List[BaseMessage]:
-    if isinstance(message, BaseMessage):
-        return [message]
-    elif isinstance(message, FunctionResult):
-        return [FunctionMessage(name=message.name, content=json.dumps(message.result))]
-    else:
-        return []
+def prompt_generator(log: MessageLog) -> List[BaseMessage]:
+    """Generate a prompt from a log of message like objects."""
+    messages = []
+    for message in log.messages:
+        if isinstance(message, BaseMessage):
+            messages.append(message)
+        elif isinstance(message, FunctionResult):
+            messages.append(
+                FunctionMessage(name=message.name, content=json.dumps(message.result))
+            )
+        else:
+            pass
+    return messages
 
 
 class OpenAIAgent:
@@ -63,7 +67,7 @@ class OpenAIAgent:
         """Initialize the chat automaton."""
         self.llm_program = create_llm_program(
             llm,
-            prompt_generator=MessageLogPromptValue.from_message_log,
+            prompt_generator=prompt_generator,
             tools=tools,
             parser=OpenAIFunctionsParser(),
         )

libs/langchain/langchain/automaton/runnables.py

@@ -44,7 +44,9 @@ def create_tool_invoker(
 
 def create_llm_program(
     llm: BaseLanguageModel,
-    prompt_generator: Callable[[MessageLog], PromptValue],
+    prompt_generator: Callable[
+        [MessageLog], Union[str, PromptValue, Sequence[BaseMessage]]
+    ],
     *,
     tools: Optional[Sequence[BaseTool]] = None,
     stop: Optional[Sequence[str]] = None,
@@ -91,6 +93,8 @@ def create_llm_program(
                     )
                 messages.append(parsed_result)
 
+        if not messages:
+            raise AssertionError(f"Expected at least one message")
         last_message = messages[-1]
 
         if tool_invoker and isinstance(last_message, FunctionCall):