gpt4free/g4f/Provider/AiAsk.py

44 lines
1.6 KiB
Python
Raw Normal View History

from __future__ import annotations
from aiohttp import ClientSession
from ..typing import AsyncGenerator
from .base_provider import AsyncGeneratorProvider
class AiAsk(AsyncGeneratorProvider):
url = "https://e.aiask.me"
supports_gpt_35_turbo = True
working = True
@classmethod
async def create_async_generator(
cls,
model: str,
messages: list[dict[str, str]],
**kwargs
) -> AsyncGenerator:
headers = {
"accept": "application/json, text/plain, */*",
"origin": cls.url,
"referer": f"{cls.url}/chat",
}
async with ClientSession(headers=headers) as session:
data = {
"continuous": True,
"id": "fRMSQtuHl91A4De9cCvKD",
"list": messages,
"models": "0",
"prompt": "",
"temperature": kwargs.get("temperature", 0.5),
"title": "",
}
buffer = ""
rate_limit = "您的免费额度不够使用这个模型啦,请点击右上角登录继续使用!"
async with session.post(f"{cls.url}/v1/chat/gpt/", json=data) as response:
response.raise_for_status()
async for chunk in response.content.iter_any():
buffer += chunk.decode()
if not rate_limit.startswith(buffer):
yield buffer
buffer = ""
elif buffer == rate_limit:
raise RuntimeError("Rate limit reached")