回答

我们以qwen为例，qwen也是兼容openai协议的，搭建一个api代理后，访问本地localhost就能流式输出结果。

点击查看python代码

# main.py
import os
import time
import hashlib
import json
from typing import List, Optional

import httpx
from fastapi import FastAPI, Request, HTTPException, Depends
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from starlette.responses import StreamingResponse


API_KEY ="千问的apikey，去百炼申请"
BASE_URL ="https://dashscope.aliyuncs.com/compatible-mode/v1"


# --- 2. Pydantic 模型 (与OpenAI API兼容) ---
# 复用之前的模型，因为我们的代理需要理解请求结构
class ChatMessage(BaseModel):
    role: str
    content: str

class ChatCompletionRequest(BaseModel):
    model: str
    messages: List[ChatMessage]
    stream: Optional[bool] = False
    # 其他OpenAI参数...
    temperature: Optional[float] = 0.7



# --- 3. FastAPI 应用和代理端点 ---
app = FastAPI(title="OpenAI-Compatible Proxy")

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

# 使用 httpx.AsyncClient 来创建可复用的、支持HTTP/2的客户端
# 这是进行异步API调用的推荐方式
client = httpx.AsyncClient(base_url=BASE_URL)

@app.post("/v1/chat/completions", dependencies=[Depends(verify_signature)])
async def chat_completions_proxy(request: Request):
  


    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {API_KEY}",
        "Accept": request.headers.get("Accept", "text/event-stream"),
    }
    
    body = await request.body()
    
    # 定义正确的端点路径
    endpoint_path = "/chat/completions"

    try:
        # 使用正确的端点路径，而不是基地址
        backend_req = client.build_request(
            method="POST",
            url=endpoint_path, 
            headers=headers,
            content=body,
            timeout=300.0
        )
        
        # 【调试技巧】: 在这里打印出将要发送到后端的信息
        print("--- Forwarding to Backend ---")
        print(f"URL: {backend_req.url}")
        print(f"Headers: {backend_req.headers}")
        print(f"Body: {backend_req.content.decode()}")
        print("---------------------------")

        backend_resp = await client.send(backend_req, stream=True)

    except httpx.RequestError as e:
        raise HTTPException(status_code=503, detail=f"后端服务请求失败: {e}")

    if backend_resp.status_code != 200:
        error_content = await backend_resp.aread()
        # 打印后端返回的原始错误，方便调试
        print(f"Backend Error ({backend_resp.status_code}): {error_content.decode()}")
        raise HTTPException(
            status_code=backend_resp.status_code, # 将后端的错误码透传给客户端
            detail=f"后端服务返回错误: {error_content.decode()}"
        )

    return StreamingResponse(
        backend_resp.aiter_bytes(),
        status_code=backend_resp.status_code,
        media_type=backend_resp.headers.get("Content-Type"),
        background=backend_resp.aclose
    )

# --- 5. 运行服务器 ---
if __name__ == "__main__":
    import uvicorn
    uvicorn.run("chatproxy:app", host="127.0.0.1", port=8100, reload=False)