langchain/chat_models/ernie.py

import json
import logging
import threading
from typing import Any, Dict, List, Mapping, Optional

import requests
from langchain_core.messages import (
    AIMessage,
    BaseMessage,
    ChatMessage,
    HumanMessage,
)
from langchain_core.outputs import ChatGeneration, ChatResult
from langchain_core.pydantic_v1 import root_validator

from langchain.callbacks.manager import CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.utils import get_from_dict_or_env

logger = logging.getLogger(__name__)


def _convert_message_to_dict(message: BaseMessage) -> dict:
    if isinstance(message, ChatMessage):
        message_dict = {"role": message.role, "content": message.content}
    elif isinstance(message, HumanMessage):
        message_dict = {"role": "user", "content": message.content}
    elif isinstance(message, AIMessage):
        message_dict = {"role": "assistant", "content": message.content}
    else:
        raise ValueError(f"Got unknown type {message}")
    return message_dict


class ErnieBotChat(BaseChatModel):
    """`ERNIE-Bot` large language model.

    ERNIE-Bot is a large language model developed by Baidu,
    covering a huge amount of Chinese data.

    To use, you should have the `ernie_client_id` and `ernie_client_secret` set,
    or set the environment variable `ERNIE_CLIENT_ID` and `ERNIE_CLIENT_SECRET`.

    Note:
    access_token will be automatically generated based on client_id and client_secret,
    and will be regenerated after expiration (30 days).

    Default model is `ERNIE-Bot-turbo`,
    currently supported models are `ERNIE-Bot-turbo`, `ERNIE-Bot`

    Example:
        .. code-block:: python

            from langchain.chat_models import ErnieBotChat
            chat = ErnieBotChat(model_name='ERNIE-Bot')

    """

    ernie_api_base: Optional[str] = None
    """Baidu application custom endpoints"""

    ernie_client_id: Optional[str] = None
    """Baidu application client id"""

    ernie_client_secret: Optional[str] = None
    """Baidu application client secret"""

    access_token: Optional[str] = None
    """access token is generated by client id and client secret, 
    setting this value directly will cause an error"""

    model_name: str = "ERNIE-Bot-turbo"
    """model name of ernie, default is `ERNIE-Bot-turbo`.
      Currently supported `ERNIE-Bot-turbo`, `ERNIE-Bot`"""

    request_timeout: Optional[int] = 60
    """request timeout for chat http requests"""

    streaming: Optional[bool] = False
    """streaming mode. not supported yet."""

    top_p: Optional[float] = 0.8
    temperature: Optional[float] = 0.95
    penalty_score: Optional[float] = 1

    _lock = threading.Lock()

    @root_validator()
    def validate_environment(cls, values: Dict) -> Dict:
        values["ernie_api_base"] = get_from_dict_or_env(
            values, "ernie_api_base", "ERNIE_API_BASE", "https://aip.baidubce.com"
        )
        values["ernie_client_id"] = get_from_dict_or_env(
            values,
            "ernie_client_id",
            "ERNIE_CLIENT_ID",
        )
        values["ernie_client_secret"] = get_from_dict_or_env(
            values,
            "ernie_client_secret",
            "ERNIE_CLIENT_SECRET",
        )
        return values

    def _chat(self, payload: object) -> dict:
        base_url = f"{self.ernie_api_base}/rpc/2.0/ai_custom/v1/wenxinworkshop/chat"
        model_paths = {
            "ERNIE-Bot-turbo": "eb-instant",
            "ERNIE-Bot": "completions",
            "ERNIE-Bot-4": "completions_pro",
            "BLOOMZ-7B": "bloomz_7b1",
            "Llama-2-7b-chat": "llama_2_7b",
            "Llama-2-13b-chat": "llama_2_13b",
            "Llama-2-70b-chat": "llama_2_70b",
        }
        if self.model_name in model_paths:
            url = f"{base_url}/{model_paths[self.model_name]}"
        else:
            raise ValueError(f"Got unknown model_name {self.model_name}")

        resp = requests.post(
            url,
            timeout=self.request_timeout,
            headers={
                "Content-Type": "application/json",
            },
            params={"access_token": self.access_token},
            json=payload,
        )
        return resp.json()

    def _refresh_access_token_with_lock(self) -> None:
        with self._lock:
            logger.debug("Refreshing access token")
            base_url: str = f"{self.ernie_api_base}/oauth/2.0/token"
            resp = requests.post(
                base_url,
                timeout=10,
                headers={
                    "Content-Type": "application/json",
                    "Accept": "application/json",
                },
                params={
                    "grant_type": "client_credentials",
                    "client_id": self.ernie_client_id,
                    "client_secret": self.ernie_client_secret,
                },
            )
            self.access_token = str(resp.json().get("access_token"))

    def _generate(
        self,
        messages: List[BaseMessage],
        stop: Optional[List[str]] = None,
        run_manager: Optional[CallbackManagerForLLMRun] = None,
        **kwargs: Any,
    ) -> ChatResult:
        if self.streaming:
            raise ValueError("`streaming` option currently unsupported.")

        if not self.access_token:
            self._refresh_access_token_with_lock()
        payload = {
            "messages": [_convert_message_to_dict(m) for m in messages],
            "top_p": self.top_p,
            "temperature": self.temperature,
            "penalty_score": self.penalty_score,
            **kwargs,
        }
        logger.debug(f"Payload for ernie api is {payload}")
        resp = self._chat(payload)
        if resp.get("error_code"):
            if resp.get("error_code") == 111:
                logger.debug("access_token expired, refresh it")
                self._refresh_access_token_with_lock()
                resp = self._chat(payload)
            else:
                raise ValueError(f"Error from ErnieChat api response: {resp}")
        return self._create_chat_result(resp)

    def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
        if "function_call" in response:
            fc_str = '{{"function_call": {}}}'.format(
                json.dumps(response.get("function_call"))
            )
            generations = [ChatGeneration(message=AIMessage(content=fc_str))]
        else:
            generations = [
                ChatGeneration(message=AIMessage(content=response.get("result")))
            ]
        token_usage = response.get("usage", {})
        llm_output = {"token_usage": token_usage, "model_name": self.model_name}
        return ChatResult(generations=generations, llm_output=llm_output)

    @property
    def _llm_type(self) -> str:
        return "ernie-bot-chat"