2023-09-03 04:26:26 -04:00
|
|
|
from __future__ import annotations
|
|
|
|
|
2024-03-10 21:41:59 -04:00
|
|
|
import json
|
2024-03-25 16:06:51 -04:00
|
|
|
import requests
|
2024-01-23 17:48:11 -05:00
|
|
|
from aiohttp import ClientSession, BaseConnector
|
2023-08-27 19:43:45 -04:00
|
|
|
|
2024-01-21 21:38:11 -05:00
|
|
|
from ..typing import AsyncResult, Messages
|
2024-04-06 18:01:04 -04:00
|
|
|
from ..requests.raise_for_status import raise_for_status
|
2024-01-23 13:44:48 -05:00
|
|
|
from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
|
2024-03-10 21:41:59 -04:00
|
|
|
from .helper import format_prompt, get_connector
|
2023-08-27 19:43:45 -04:00
|
|
|
|
2024-01-23 13:44:48 -05:00
|
|
|
class HuggingChat(AsyncGeneratorProvider, ProviderModelMixin):
|
2023-09-23 05:42:30 -04:00
|
|
|
url = "https://huggingface.co/chat"
|
2023-08-27 19:43:45 -04:00
|
|
|
working = True
|
2024-01-22 23:02:14 -05:00
|
|
|
default_model = "meta-llama/Llama-2-70b-chat-hf"
|
|
|
|
models = [
|
|
|
|
"mistralai/Mixtral-8x7B-Instruct-v0.1",
|
2024-03-25 16:06:51 -04:00
|
|
|
"google/gemma-7b-it",
|
2024-01-22 23:02:14 -05:00
|
|
|
"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",
|
2024-02-06 01:29:30 -05:00
|
|
|
"openchat/openchat-3.5-0106",
|
2024-01-22 23:02:14 -05:00
|
|
|
]
|
2024-01-23 13:44:48 -05:00
|
|
|
model_aliases = {
|
2024-03-25 16:06:51 -04:00
|
|
|
"openchat/openchat_3.5": "openchat/openchat-3.5-0106",
|
2024-01-22 23:02:14 -05:00
|
|
|
}
|
2023-08-27 19:43:45 -04:00
|
|
|
|
2024-03-25 16:06:51 -04:00
|
|
|
@classmethod
|
|
|
|
def get_models(cls):
|
|
|
|
if not cls.models:
|
|
|
|
url = f"{cls.url}/__data.json"
|
|
|
|
data = requests.get(url).json()["nodes"][0]["data"]
|
|
|
|
models = [data[key]["name"] for key in data[data[0]["models"]]]
|
|
|
|
cls.models = [data[key] for key in models]
|
|
|
|
return cls.models
|
|
|
|
|
2023-08-27 19:43:45 -04:00
|
|
|
@classmethod
|
|
|
|
async def create_async_generator(
|
|
|
|
cls,
|
|
|
|
model: str,
|
2023-10-09 07:33:20 -04:00
|
|
|
messages: Messages,
|
2023-08-27 19:43:45 -04:00
|
|
|
stream: bool = True,
|
|
|
|
proxy: str = None,
|
2024-01-23 17:48:11 -05:00
|
|
|
connector: BaseConnector = None,
|
2023-12-06 03:35:36 -05:00
|
|
|
web_search: bool = False,
|
2023-08-27 19:43:45 -04:00
|
|
|
cookies: dict = None,
|
|
|
|
**kwargs
|
2023-10-09 07:33:20 -04:00
|
|
|
) -> AsyncResult:
|
2024-03-10 21:41:59 -04:00
|
|
|
options = {"model": cls.get_model(model)}
|
|
|
|
system_prompt = "\n".join([message["content"] for message in messages if message["role"] == "system"])
|
|
|
|
if system_prompt:
|
|
|
|
options["preprompt"] = system_prompt
|
|
|
|
messages = [message for message in messages if message["role"] != "system"]
|
2023-08-27 19:43:45 -04:00
|
|
|
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,
|
2024-01-23 17:48:11 -05:00
|
|
|
headers=headers,
|
2024-01-23 18:46:35 -05:00
|
|
|
connector=get_connector(connector, proxy)
|
2023-08-27 19:43:45 -04:00
|
|
|
) as session:
|
2024-04-06 18:01:04 -04:00
|
|
|
async with session.post(f"{cls.url}/conversation", json=options) as response:
|
|
|
|
await raise_for_status(response)
|
2023-08-27 19:43:45 -04:00
|
|
|
conversation_id = (await response.json())["conversationId"]
|
2024-03-10 21:41:59 -04:00
|
|
|
async with session.get(f"{cls.url}/conversation/{conversation_id}/__data.json") as response:
|
2024-04-06 18:01:04 -04:00
|
|
|
await raise_for_status(response)
|
2024-03-10 21:41:59 -04:00
|
|
|
data: list = (await response.json())["nodes"][1]["data"]
|
|
|
|
keys: list[int] = data[data[0]["messages"]]
|
|
|
|
message_keys: dict = data[keys[0]]
|
|
|
|
message_id: str = data[message_keys["id"]]
|
|
|
|
options = {
|
|
|
|
"id": message_id,
|
2023-08-27 19:43:45 -04:00
|
|
|
"inputs": format_prompt(messages),
|
2024-03-10 21:41:59 -04:00
|
|
|
"is_continue": False,
|
2023-10-01 00:38:11 -04:00
|
|
|
"is_retry": False,
|
2023-12-06 03:35:36 -05:00
|
|
|
"web_search": web_search
|
2023-08-27 19:43:45 -04:00
|
|
|
}
|
2024-03-10 21:41:59 -04:00
|
|
|
async with session.post(f"{cls.url}/conversation/{conversation_id}", json=options) as response:
|
2023-12-15 17:58:13 -05:00
|
|
|
first_token = True
|
2023-10-01 00:38:11 -04:00
|
|
|
async for line in response.content:
|
2024-04-06 18:01:04 -04:00
|
|
|
await raise_for_status(response)
|
2024-03-10 21:41:59 -04:00
|
|
|
line = json.loads(line)
|
2023-10-01 00:38:11 -04:00
|
|
|
if "type" not in line:
|
|
|
|
raise RuntimeError(f"Response: {line}")
|
|
|
|
elif line["type"] == "stream":
|
2023-12-15 17:58:13 -05:00
|
|
|
token = line["token"]
|
|
|
|
if first_token:
|
|
|
|
token = token.lstrip()
|
|
|
|
first_token = False
|
|
|
|
yield token
|
2023-10-01 00:38:11 -04:00
|
|
|
elif line["type"] == "finalAnswer":
|
|
|
|
break
|
2024-04-06 18:01:04 -04:00
|
|
|
async with session.delete(f"{cls.url}/conversation/{conversation_id}") as response:
|
|
|
|
await raise_for_status(response)
|