gpt4free/g4f/Provider/HuggingChat.py

from __future__ import annotations

import json, uuid

from aiohttp import ClientSession

from ..typing import AsyncResult, Messages
from .base_provider import AsyncGeneratorProvider
from .helper import format_prompt, get_cookies


class HuggingChat(AsyncGeneratorProvider):
    url = "https://huggingface.co/chat"
    working = True
    default_model = "meta-llama/Llama-2-70b-chat-hf"
    models = [
        "mistralai/Mixtral-8x7B-Instruct-v0.1",
        "meta-llama/Llama-2-70b-chat-hf",
        "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
        "codellama/CodeLlama-34b-Instruct-hf",
        "mistralai/Mistral-7B-Instruct-v0.2",
        "openchat/openchat-3.5-0106"
    ]
    model_map = {
        "openchat/openchat_3.5": "openchat/openchat-3.5-1210",
        "mistralai/Mixtral-8x7B-Instruct-v0.1": "mistralai/Mistral-7B-Instruct-v0.2"
    }

    @classmethod
    async def create_async_generator(
        cls,
        model: str,
        messages: Messages,
        stream: bool = True,
        proxy: str = None,
        web_search: bool = False,
        cookies: dict = None,
        **kwargs
    ) -> AsyncResult:
        if not model:
            model = cls.default_model
        elif model in cls.model_map:
            model = cls.model_map[model]
        elif model not in cls.models:
            raise ValueError(f"Model is not supported: {model}")
        if not cookies:
            cookies = get_cookies(".huggingface.co")

        headers = {
            'User-Agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/111.0.0.0 Safari/537.36',
        }
        async with ClientSession(
            cookies=cookies,
            headers=headers
        ) as session:
            async with session.post(f"{cls.url}/conversation", json={"model": model}, proxy=proxy) as response:
                conversation_id = (await response.json())["conversationId"]

            send = {
                "id": str(uuid.uuid4()),
                "inputs": format_prompt(messages),
                "is_retry": False,
                "response_id": str(uuid.uuid4()),
                "web_search": web_search
            }
            async with session.post(f"{cls.url}/conversation/{conversation_id}", json=send, proxy=proxy) as response:
                first_token = True
                async for line in response.content:
                    line = json.loads(line[:-1])
                    if "type" not in line:
                        raise RuntimeError(f"Response: {line}")
                    elif line["type"] == "stream":
                        token = line["token"]
                        if first_token:
                            token = token.lstrip()
                            first_token = False
                        yield token
                    elif line["type"] == "finalAnswer":
                        break
                
            async with session.delete(f"{cls.url}/conversation/{conversation_id}", proxy=proxy) as response:
                response.raise_for_status()