llm.py

from langchain.llms.base import LLM
from typing import Optional, List
from langchain.llms.utils import enforce_stop_tokens
import requests

FORWARD_KEY = 'fk198719-Pmvv22OqZiovaxRq6YxCzkTcd6UVVX5O'

class ChatGLM(LLM):
    max_length: int = 10000
    temperature: float = 0
    top_p = 0.9
    tokenizer: object = None
    model: object = None
    history_len: int = 10
    history = []
    URL = 'http://183.131.3.48:9200'
    HEADERS = {'Content-Type': 'application/json'}

    @property
    def _llm_type(self) -> str:
        return "ChatGLM"

    def _call(self,
              prompt: str,
              history:  Optional[List[List[str]]] = None,
              stop: Optional[List[str]] = None) -> str:

        if history:
            history = [i for i in history if i[0] is not None]  # clear out the system message
            history = history[-self.history_len:]

        params = {'tokenizers': self.tokenizer, 'prompt': prompt, 'history': history, 'top_p': self.top_p,
                  'max_length': self.max_length, 'temperature': self.temperature}

        response = requests.post(self.URL, headers=self.HEADERS, json=params).json()
        answer = response['response']
        if stop is not None:
            answer = enforce_stop_tokens(answer, stop)
        # question = prompt.split('question:\n')[-1]
        # self.history = self.history+[[prompt, response]]

        return answer


class OpenAI3(LLM):
    max_length: int = 10000
    temperature: float = 0.2
    top_p = 0.9
    tokenizer: object = None
    model: object = None
    history_len: int = 10
    history = []
    HEADERS = {'Content-Type': 'application/json', 'Authorization': 'Bearer fk198719-Pmvv22OqZiovaxRq6YxCzkTcd6UVVX5O'}
    URL ='https://openai.api2d.net/v1/chat/completions'
    MODEL_NAME = "gpt-3.5-turbo"


    @property
    def _llm_type(self) -> str:
        return "ChatGLM"

    def _call(self,
              prompt: str,
              history: Optional[List[List[str]]] = None,
              stop: Optional[List[str]] = None) -> str:

        if history:
            history = [i for i in history if i[0] is not None]
            history = history[-self.history_len:]
            message = [[{"role": "user", "content": i[0]}, {"role": "assistant", "content": i[1]}] for i in history]
            message = sum(message, [])
        else:
            message = []
        message.append({"role": "user", "content": prompt})
        params = {"model": self.MODEL_NAME, "messages": message, 'temperature': self.temperature}
        response = requests.post(self.URL, headers=self.HEADERS, json=params).json()
        answer = response['choices'][0]['message']['content']
        if stop is not None:
            answer = enforce_stop_tokens(answer, stop)
        # question = prompt.split('question:\n')[-1]
        # self.history = self.history+[[question, response]]

        return answer


class OpenAI4(LLM):
    max_length: int = 10000
    temperature: float = 0.2
    top_p = 0.9
    tokenizer: object = None
    model: object = None
    history_len: int = 10
    history = []
    HEADERS = {'Content-Type': 'application/json', 'Authorization': 'Bearer fk198719-Pmvv22OqZiovaxRq6YxCzkTcd6UVVX5O'}
    URL ='https://openai.api2d.net/v1/chat/completions'
    MODEL_NAME = "gpt-4"


    @property
    def _llm_type(self) -> str:
        return "ChatGLM"

    def _call(self,
              prompt: str,
              history: Optional[List[List[str]]] = None,
              stop: Optional[List[str]] = None) -> str:

        if history:
            history = [i for i in history if i[0] is not None]
            history = history[-self.history_len:]
            message = [[{"role": "user", "content": i[0]}, {"role": "assistant", "content": i[1]}] for i in history]
            message = sum(message, [])
        else:
            message = []
        message.append({"role": "user", "content": prompt})
        params = {"model": self.MODEL_NAME, "messages": message, 'temperature': self.temperature}
        response = requests.post(self.URL, headers=self.HEADERS, json=params).json()
        answer = response['choices'][0]['message']['content']
        if stop is not None:
            answer = enforce_stop_tokens(answer, stop)
        # question = prompt.split('question:\n')[-1]
        # self.history = self.history+[[question, response]]

        return answer