innolux/app.py

import openai
import uvicorn

from typing import List
from pydantic import BaseModel
from fastapi import FastAPI, Body
from fastapi.middleware.cors import CORSMiddleware
from langchain.schema import SystemMessage, AIMessage, HumanMessage
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain.callbacks import get_openai_callback

import os
from dotenv import load_dotenv
load_dotenv()


from supabase.client import Client, create_client
supabase_url = os.environ.get("SUPABASE_URL")
supabase_key = os.environ.get("SUPABASE_KEY")
supabase: Client = create_client(supabase_url, supabase_key)

chat_model = ChatOpenAI(model='gpt-4o-mini')

app = FastAPI()
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

from semantic_search import semantic_cache
from api import whisper
app.include_router(whisper.router, prefix='/whisper', tags=['whisper'])


class ChatHistoryItem(BaseModel):
    q: str
    a: str

@app.post("/chat/")
async def chat(message, chat_history: List[ChatHistoryItem] = Body(...)):
    print(chat_history)
    messages_list = [
        SystemMessage(content="你是一名日照中心的志工，你的職責是陪伴老人聊天，你需要考量到老人的健康與安全，並能安撫老人，只有家人准許時才能離開日照中心。請用繁體中文"),
        AIMessage(content="你好！很高興能和您聊天。今天您過得怎麼樣呢？有沒有什麼想分享的事情？")
    ]
    
    for item in chat_history[-10:]:
        if item.q == "string" or item.a == "string" : continue
        messages_list.append(HumanMessage(content=item.q))
        messages_list.append(AIMessage(content=item.a))

    messages_list.append(HumanMessage(content=message))
    print(messages_list)
    prompt = ChatPromptTemplate(
        messages=messages_list
    )
    try:
        with get_openai_callback() as cb:
            # cache_question, cache_answer = semantic_cache(supabase, message)
            cache_question, cache_answer, video_cache = semantic_cache(supabase, message, SIMILARITY_THRESHOLD=0.83)
            if cache_answer:
                save_history(message, cache_answer)
                # return {"message": cache_answer}
                if video_cache is not None:
                    return {"message": cache_answer, "video_cache": video_cache}
                else:
                    return {"message": cache_answer}

            AIMessage_ = chat_model.invoke(prompt.format_messages()).content
            save_history(message, AIMessage_)
            
    except openai.RateLimitError as e:
            print(f"Rate limit exceeded: {e}")
            return {'message': "Current quota exceeded."}
    
    return {"message": AIMessage_}

def save_history(question, answer):
    response = (
            supabase.table("INNOLUX_record")
            .insert({"question": question, "answer": answer})
            .execute()
        )

from starlette.responses import JSONResponse
@app.get("/health") # 使用網址給 kuma 監測
async def health_check():
    return JSONResponse(content={"status": "ok"}, status_code=200)


if __name__ == "__main__":
    
    try:
        uvicorn.run("app:app", reload=False, port=8087, host='cmm.ai', ssl_keyfile="/etc/letsencrypt/live/cmm.ai/privkey.pem", ssl_certfile="/etc/letsencrypt/live/cmm.ai/fullchain.pem")
    finally:
        import shutil
        cache_db = "chroma_db"
        shutil.rmtree(cache_db)