react_single_input.py — langchain Source File

Source Code

import re

from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.exceptions import OutputParserException
from typing_extensions import override

from langchain_classic.agents.agent import AgentOutputParser
from langchain_classic.agents.mrkl.prompt import FORMAT_INSTRUCTIONS

FINAL_ANSWER_ACTION = "Final Answer:"
MISSING_ACTION_AFTER_THOUGHT_ERROR_MESSAGE = (
    "Invalid Format: Missing 'Action:' after 'Thought:'"
)
MISSING_ACTION_INPUT_AFTER_ACTION_ERROR_MESSAGE = (
    "Invalid Format: Missing 'Action Input:' after 'Action:'"
)
FINAL_ANSWER_AND_PARSABLE_ACTION_ERROR_MESSAGE = (
    "Parsing LLM output produced both a final answer and a parse-able action:"
)


class ReActSingleInputOutputParser(AgentOutputParser):
    """Parses ReAct-style LLM calls that have a single tool input.

    Expects output to be in one of two formats.

    If the output signals that an action should be taken,
    should be in the below format. This will result in an AgentAction
    being returned.

    ```
    Thought: agent thought here
    Action: search
    Action Input: what is the temperature in SF?
    ```

    If the output signals that a final answer should be given,
    should be in the below format. This will result in an AgentFinish
    being returned.

    ```
    Thought: agent thought here
    Final Answer: The temperature is 100 degrees
    ```

    """

    @override
    def get_format_instructions(self) -> str:
        return FORMAT_INSTRUCTIONS

    @override
    def parse(self, text: str) -> AgentAction | AgentFinish:
        includes_answer = FINAL_ANSWER_ACTION in text
        regex = (
            r"Action\s*\d*\s*:[\s]*(.*?)[\s]*Action\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
        )
        action_match = re.search(regex, text, re.DOTALL)
        if action_match:
            if includes_answer:
                msg = f"{FINAL_ANSWER_AND_PARSABLE_ACTION_ERROR_MESSAGE}: {text}"
                raise OutputParserException(msg)
            action = action_match.group(1).strip()
            action_input = action_match.group(2)
            tool_input = action_input.strip(" ")
            tool_input = tool_input.strip('"')

            return AgentAction(action, tool_input, text)

        if includes_answer:
            return AgentFinish(
                {"output": text.rsplit(FINAL_ANSWER_ACTION, maxsplit=1)[-1].strip()},
                text,
            )

        if not re.search(r"Action\s*\d*\s*:[\s]*(.*?)", text, re.DOTALL):
            msg = f"Could not parse LLM output: `{text}`"
            raise OutputParserException(
                msg,
                observation=MISSING_ACTION_AFTER_THOUGHT_ERROR_MESSAGE,
                llm_output=text,
                send_to_llm=True,
            )
        if not re.search(
            r"[\s]*Action\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)",
            text,
            re.DOTALL,
        ):
            msg = f"Could not parse LLM output: `{text}`"
            raise OutputParserException(
                msg,
                observation=MISSING_ACTION_INPUT_AFTER_ACTION_ERROR_MESSAGE,
                llm_output=text,
                send_to_llm=True,
            )
        msg = f"Could not parse LLM output: `{text}`"
        raise OutputParserException(msg)

    @property
    def _type(self) -> str:
        return "react-single-input"