2023-07-27 23:22:07 +08:00
|
|
|
from fastapi import Body
|
|
|
|
|
from fastapi.responses import StreamingResponse
|
|
|
|
|
from configs.model_config import llm_model_dict, LLM_MODEL
|
2023-08-06 18:32:10 +08:00
|
|
|
from server.chat.utils import wrap_done
|
2023-07-27 23:22:07 +08:00
|
|
|
from langchain.chat_models import ChatOpenAI
|
|
|
|
|
from langchain import LLMChain
|
|
|
|
|
from langchain.callbacks import AsyncIteratorCallbackHandler
|
|
|
|
|
from typing import AsyncIterable
|
|
|
|
|
import asyncio
|
2023-08-08 23:54:51 +08:00
|
|
|
from langchain.prompts.chat import ChatPromptTemplate
|
2023-08-10 21:26:05 +08:00
|
|
|
from typing import List
|
2023-08-08 23:54:51 +08:00
|
|
|
from server.chat.utils import History
|
2023-07-27 23:22:07 +08:00
|
|
|
|
|
|
|
|
|
2023-08-09 18:15:14 +08:00
|
|
|
def chat(query: str = Body(..., description="用户输入", examples=["恼羞成怒"]),
|
2023-08-09 10:48:37 +08:00
|
|
|
history: List[History] = Body([],
|
2023-08-10 21:26:05 +08:00
|
|
|
description="历史对话",
|
|
|
|
|
examples=[[
|
|
|
|
|
{"role": "user", "content": "我们来玩成语接龙,我先来,生龙活虎"},
|
|
|
|
|
{"role": "assistant", "content": "虎头虎脑"}]]
|
|
|
|
|
),
|
2023-08-08 23:54:51 +08:00
|
|
|
):
|
2023-08-09 12:09:45 +08:00
|
|
|
history = [History(**h) if isinstance(h, dict) else h for h in history]
|
2023-08-10 21:26:05 +08:00
|
|
|
|
2023-08-08 23:54:51 +08:00
|
|
|
async def chat_iterator(query: str,
|
2023-08-09 10:48:37 +08:00
|
|
|
history: List[History] = [],
|
2023-08-08 23:54:51 +08:00
|
|
|
) -> AsyncIterable[str]:
|
2023-07-27 23:22:07 +08:00
|
|
|
callback = AsyncIteratorCallbackHandler()
|
|
|
|
|
|
|
|
|
|
model = ChatOpenAI(
|
|
|
|
|
streaming=True,
|
|
|
|
|
verbose=True,
|
|
|
|
|
callbacks=[callback],
|
|
|
|
|
openai_api_key=llm_model_dict[LLM_MODEL]["api_key"],
|
|
|
|
|
openai_api_base=llm_model_dict[LLM_MODEL]["api_base_url"],
|
|
|
|
|
model_name=LLM_MODEL
|
|
|
|
|
)
|
|
|
|
|
|
2023-08-08 23:54:51 +08:00
|
|
|
chat_prompt = ChatPromptTemplate.from_messages(
|
|
|
|
|
[i.to_msg_tuple() for i in history] + [("human", "{input}")])
|
|
|
|
|
chain = LLMChain(prompt=chat_prompt, llm=model)
|
2023-07-27 23:22:07 +08:00
|
|
|
|
|
|
|
|
# Begin a task that runs in the background.
|
|
|
|
|
task = asyncio.create_task(wrap_done(
|
2023-08-08 23:54:51 +08:00
|
|
|
chain.acall({"input": query}),
|
2023-07-27 23:22:07 +08:00
|
|
|
callback.done),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
async for token in callback.aiter():
|
|
|
|
|
# Use server-sent-events to stream the response
|
|
|
|
|
yield token
|
|
|
|
|
await task
|
2023-08-08 23:54:51 +08:00
|
|
|
|
|
|
|
|
return StreamingResponse(chat_iterator(query, history),
|
|
|
|
|
media_type="text/event-stream")
|