2023-09-03 11:26:26 +03:00
|
|
|
from __future__ import annotations
|
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
import json
|
|
|
|
import uuid
|
2023-09-03 11:26:26 +03:00
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
from aiohttp import ClientSession
|
2023-07-28 13:07:17 +03:00
|
|
|
|
2023-10-15 20:10:25 +03:00
|
|
|
from ...typing import AsyncResult, Messages
|
|
|
|
from ..base_provider import AsyncGeneratorProvider, format_prompt
|
2023-07-28 13:07:17 +03:00
|
|
|
|
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
class H2o(AsyncGeneratorProvider):
|
|
|
|
url = "https://gpt-gm.h2o.ai"
|
2023-09-05 18:27:24 +03:00
|
|
|
model = "h2oai/h2ogpt-gm-oasst1-en-2048-falcon-40b-v1"
|
2023-07-28 13:07:17 +03:00
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
@classmethod
|
|
|
|
async def create_async_generator(
|
|
|
|
cls,
|
2023-07-28 13:07:17 +03:00
|
|
|
model: str,
|
2023-10-10 10:49:29 +03:00
|
|
|
messages: Messages,
|
2023-08-28 02:43:45 +03:00
|
|
|
proxy: str = None,
|
|
|
|
**kwargs
|
2023-10-10 10:49:29 +03:00
|
|
|
) -> AsyncResult:
|
2023-08-28 02:43:45 +03:00
|
|
|
model = model if model else cls.model
|
2023-10-23 10:46:25 +03:00
|
|
|
headers = {"Referer": f"{cls.url}/"}
|
2023-07-28 13:07:17 +03:00
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
async with ClientSession(
|
|
|
|
headers=headers
|
|
|
|
) as session:
|
|
|
|
data = {
|
|
|
|
"ethicsModalAccepted": "true",
|
|
|
|
"shareConversationsWithModelAuthors": "true",
|
|
|
|
"ethicsModalAcceptedAt": "",
|
|
|
|
"activeModel": model,
|
|
|
|
"searchEnabled": "true",
|
|
|
|
}
|
|
|
|
async with session.post(
|
2023-09-20 07:12:34 +03:00
|
|
|
f"{cls.url}/settings",
|
2023-08-28 02:43:45 +03:00
|
|
|
proxy=proxy,
|
|
|
|
data=data
|
|
|
|
) as response:
|
|
|
|
response.raise_for_status()
|
2023-07-28 13:07:17 +03:00
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
async with session.post(
|
2023-09-20 07:12:34 +03:00
|
|
|
f"{cls.url}/conversation",
|
2023-08-28 02:43:45 +03:00
|
|
|
proxy=proxy,
|
|
|
|
json={"model": model},
|
|
|
|
) as response:
|
|
|
|
response.raise_for_status()
|
|
|
|
conversationId = (await response.json())["conversationId"]
|
2023-07-28 13:07:17 +03:00
|
|
|
|
2023-08-28 02:43:45 +03:00
|
|
|
data = {
|
|
|
|
"inputs": format_prompt(messages),
|
|
|
|
"parameters": {
|
|
|
|
"temperature": 0.4,
|
|
|
|
"truncate": 2048,
|
|
|
|
"max_new_tokens": 1024,
|
|
|
|
"do_sample": True,
|
|
|
|
"repetition_penalty": 1.2,
|
|
|
|
"return_full_text": False,
|
|
|
|
**kwargs
|
|
|
|
},
|
|
|
|
"stream": True,
|
|
|
|
"options": {
|
|
|
|
"id": str(uuid.uuid4()),
|
|
|
|
"response_id": str(uuid.uuid4()),
|
|
|
|
"is_retry": False,
|
|
|
|
"use_cache": False,
|
|
|
|
"web_search_id": "",
|
|
|
|
},
|
|
|
|
}
|
|
|
|
async with session.post(
|
2023-09-20 07:12:34 +03:00
|
|
|
f"{cls.url}/conversation/{conversationId}",
|
2023-08-28 02:43:45 +03:00
|
|
|
proxy=proxy,
|
|
|
|
json=data
|
|
|
|
) as response:
|
|
|
|
start = "data:"
|
|
|
|
async for line in response.content:
|
|
|
|
line = line.decode("utf-8")
|
|
|
|
if line and line.startswith(start):
|
|
|
|
line = json.loads(line[len(start):-1])
|
|
|
|
if not line["token"]["special"]:
|
|
|
|
yield line["token"]["text"]
|
2023-07-28 13:07:17 +03:00
|
|
|
|
2023-09-20 07:12:34 +03:00
|
|
|
async with session.delete(
|
|
|
|
f"{cls.url}/conversation/{conversationId}",
|
|
|
|
proxy=proxy,
|
|
|
|
) as response:
|
2023-11-20 15:59:14 +03:00
|
|
|
response.raise_for_status()
|