以openai的gpt3 5为例的大模型流式输出实现（原始、Sanic、Flask）- 附免费的key

AI肥肥旭

于 2024-10-09 23:09:10 发布

阅读量156

点赞数 3

文章标签： gpt-3 flask python 大模型流式输出

本文链接：https://blog.csdn.net/CSDN_Lrcx/article/details/142798815

版权

以openai的gpt3.5为例的大模型流式输出实现（原始、Sanic、Flask）- 附免费的apikey水龙头

type: Post
status: Draft
date: 2024/10/09

📝 原始

import openai

openai.api_key = 'sk-xxx'
openai.base_url = 'xxx'

resp = openai.chat.completions.create(
    model = 'gpt-3.5-turbo',
    messages = [{"role":'user', "content": '详细介绍一下内马尔'}],
    stream = True
)

for chunk in resp:
    # if chunk.choices[0].delta.content:
    print(chunk.choices[0].delta.content, end='')

🤗 Sanic

import json
from sanic import Sanic, response
import openai

app = Sanic(__name__)

openai.api_key = "sk-xxx"
openai.base_url = "xxx"

async def process_large_model(prompt):
    resp = openai.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}],
        stream=True,
    )
    for chunk in resp:
        yield chunk.choices[0].delta.content

@app.post("/large_model")
async def large_model_handler(request):
    prompt = request.json.get("prompt")

    headers = {"Cache-Control": "no-cache"}

    async def stream_fn(response):
        accumulated_response = ""
        async for data in process_large_model(prompt):
            # gpt3.5 最后返回的内容是None，如果是None，则结束流式响应
            if data != None:
                # 和SSE协议的\n\n进行区分才进行的替换，如果你发现你的流式输出多了很多empty，多半就是没有替换惹的祸
                data = data.replace("\\n", "__hh__").replace("\n", "__hh__")
                accumulated_response += data
                # 最终响应的组装形式，可以根据想要输出的内容进行拼接，但必须是data：开头，\n\n结尾。
                # output = "data: " + accumulated_response + "\n\n"
                output = "data: " + json.dumps({"query":prompt, "answer":accumulated_response}, ensure_ascii=False) + "\n\n"
                await response.write(output)
            else:
                await response.eof()

    return response.ResponseStream(
        stream_fn, content_type="text/event-stream", headers=headers
    )

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=8080, auto_reload=True)

🤗 Flask

from flask import Flask, Response, request
import openai

openai.api_key = "sk-xxx"
openai.base_url = "xxx"

app = Flask(__name__)

def process_large_model(prompt):
    resp = openai.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}],
        stream=True,
    )
    accumulated_response = ""
    for chunk in resp:
        if chunk.choices[0].delta.content != None:
            accumulated_response += chunk.choices[0].delta.content
            accumulated_response = accumulated_response.replace("\\n", "__hh__").replace("\n", "__hh__")
            yield "data: " + accumulated_response + "\n\n"
        # else:
        #     yield "data: [DONE] \n\n"

@app.post("/large_model")
def large_model():
    prompt = request.json.get("prompt")
    return Response(
        process_large_model(prompt),
        # mimetype="text/event-stream",
        content_type="text/event-stream",
    )

if __name__ == "__main__":
    app.run("0.0.0.0", port=8081, debug=True)