仿照ChatGLM3部署，参考了Qwen模型的文档，模型地址https://modelscope.cn/models/qwen/Qwen1.5-1.8B-Chat/summary
http接口
- 服务端代码
api.py
from fastapi import FastAPI, Request
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
import uvicorn
import json
import datetime
import torch# 设置设备参数
DEVICE = "cuda" # 使用CUDA
DEVICE_ID = "0" # CUDA设备ID，如果未设置则为空
CUDA_DEVICE = f"{DEVICE}:{DEVICE_ID}" if DEVICE_ID else DEVICE # 组合CUDA设备信息# 加载预训练的分词器和模型
model_name_or_path = '/root/autodl-tmp/qwen/Qwen1.5-1.8B-Chat'
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(model_name_or_path, device_map="auto", torch_dtype=torch.bfloat16)# 清理GPU内存函数
def torch_gc():if torch.cuda.is_available(): # 检查是否可用CUDAwith torch.cuda.device(CUDA_DEVICE): # 指定CUDA设备torch.cuda.empty_cache() # 清空CUDA缓存torch.cuda.ipc_collect() # 收集CUDA内存碎片# 创建FastAPI应用
app = FastAPI()# 处理POST请求的端点
@app.post("/")
async def create_item(request: Request):global model, tokenizer # 声明全局变量以便在函数内部使用模型和分词器json_post_raw = await request.json() # 获取POST请求的JSON数据json_post = json.dumps(json_post_raw) # 将JSON数据转换为字符串json_post_list = json.loads(json_post) # 将字符串转换为Python对象prompt = json_post_list.get('prompt') # 获取请求中的提示messages = [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": prompt}]# 调用模型进行对话生成input_ids = tokenizer.apply_chat_template(messages,tokenize=False,add_generation_prompt=True)model_inputs = tokenizer([input_ids], return_tensors="pt").to('cuda')generated_ids = model.generate(model_inputs.input_ids,max_new_tokens=512)generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]now = datetime.datetime.now() # 获取当前时间time = now.strftime("%Y-%m-%d %H:%M:%S") # 格式化时间为字符串# 构建响应JSONanswer = {"response": response,"status": 200,"time": time}# 构建日志信息log = "[" + time + "] " + '", prompt:"' + prompt + '", response:"' + repr(response) + '"'print(log) # 打印日志torch_gc() # 执行GPU内存清理return answer # 返回响应# 主函数入口
if __name__ == '__main__':# 启动FastAPI应用# 用6006端口可以将autodl的端口映射到本地，从而在本地使用apiuvicorn.run("api:app", host='127.0.0.1', port=6006, workers=2) # 在指定端口和主机上启动应用# gunicorn api:app -w 3 -k uvicorn.workers.UvicornWorker -b 127.0.0.1:6006
- 客户端代码
clientapi.py
import requests
import jsondef get_completion(prompt):headers = {'Content-Type': 'application/json'}data = {"prompt": prompt}response = requests.post(url='http://127.0.0.1:6006', headers=headers, data=json.dumps(data))return response.json()['response']if __name__ == '__main__':print(get_completion('你可以记录之前说过的内容吗'))
- 压测代码
locusthttp.py，运行locust -f locusthttp.py（有UI），使用vmstat或top查看进程、内存、CPU等情况。
import jsonfrom locust import HttpUser, TaskSet, task# 定义用户行为
class UserBehavior(TaskSet):# 任一测试用例执行前均会执行一次def on_start(self):print('开始性能测试')# 表示一个用户为行，访问百度首页。使用 @task装饰该方法为一个事务。client.get()用于指请求的路径“ / ”，因为是百度首页，所以指定为根路径。@task(1)def index(self):self.client.get("/")@task(2) # task()参数用于指定该行为的执行权重。参数越大每次被虚拟用户执行的概率越高。如果不设置默认为1。def index2(self):headers = {'Content-Type': 'application/json'}data = {'prompt': '你知道珠穆朗玛峰吗'}self.client.post(url='/index2', headers=headers, data=json.dumps(data))@task(2) # task()参数用于指定该行为的执行权重。参数越大每次被虚拟用户执行的概率越高。如果不设置默认为1。def index3(self):headers = {'Content-Type': 'application/json'}data = {'prompt': '你是谁'}self.client.post(url='/index3', headers=headers, data=json.dumps(data))@task(2) # task()参数用于指定该行为的执行权重。参数越大每次被虚拟用户执行的概率越高。如果不设置默认为1。def index4(self):headers = {'Content-Type': 'application/json'}data = {'prompt': '西红柿炒番茄怎么做'}self.client.post(url='/index4', headers=headers, data=json.dumps(data))# 用于设置性能测试
class WebsiteUser(HttpUser):# 指向一个定义的用户行为类。tasks = [UserBehavior]# 执行事务之间用户等待时间的下界（单位：毫秒）。如果TaskSet类中有覆盖，以TaskSet 中的定义为准。min_wait = 3000# 执行事务之间用户等待时间的上界（单位：毫秒）。如果TaskSet类中有覆盖，以TaskSet中的定义为准。max_wait = 6000# 设置 Locust 多少秒后超时，如果为 None ,则不会超时。stop_timeout = 5# 一个Locust实例被挑选执行的权重，数值越大，执行频率越高。在一个 locustfile.py 文件中可以同时定义多个 HttpUser 子类，然后分配他们的执行权重weight = 3# 脚本指定host执行测试时则不在需要指定host = "http://127.0.0.1:6006"
WebSocket长连接
- 服务端代码
websocketapi.py
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
from fastapi.responses import HTMLResponse
from fastapi.middleware.cors import CORSMiddleware
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
import uvicorn
import torchpretrained = "/root/autodl-tmp/qwen/Qwen1.5-1.8B-Chat"
tokenizer = AutoTokenizer.from_pretrained(pretrained, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(pretrained, device_map="auto", torch_dtype=torch.bfloat16)
model = model.eval()
app = FastAPI()app.add_middleware(CORSMiddleware
)with open('websocket_demo.html') as f:html = f.read()@app.get("/")
async def get():return HTMLResponse(html)@app.websocket("/ws")
async def websocket_endpoint(websocket: WebSocket):"""input: JSON String of {"prompt": ""}output: JSON String of {"response": "", "status": 200}status 200 stand for response ended, else not"""await websocket.accept()try:while True:json_request = await websocket.receive_json()prompt = json_request['prompt']messages = [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": prompt}]input_ids = tokenizer.apply_chat_template(messages,tokenize=False,add_generation_prompt=True)model_inputs = tokenizer([input_ids], return_tensors="pt").to('cuda')generated_ids = model.generate(model_inputs.input_ids,max_new_tokens=512)generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]await websocket.send_json({"response": response,"status": 202,})await websocket.send_json({"status": 200})except WebSocketDisconnect:passdef main():uvicorn.run(f"{__name__}:app", host='127.0.0.1', port=6006, workers=2)if __name__ == '__main__':main()# gunicorn websocket_api_qwen:app -w 3 -k uvicorn.workers.UvicornWorker -b 127.0.0.1:6006
- 前端代码
websocket.html
<!DOCTYPE html>
<html lang="en">
<head><title>Chat</title>
</head>
<body>
<h1>WebSocket Chat</h1>
<form action="" onsubmit="return false;" id="form"><label for="messageText"></label><input type="text" id="messageText" autocomplete="off"/><button type="submit">Send</button>
</form>
<ul id='messageBox'>
</ul>
<script>let ws = new WebSocket("ws://" + location.host + "/ws");let history = [];let last_message_element = null;function appendMessage(text, sender, dom = null) {if (dom === null) {let messageBox = document.getElementById('messageBox');dom = document.createElement('li');messageBox.appendChild(dom);}dom.innerText = sender + '：' + text;return dom}function sendMessage(event) {if (last_message_element !== null) { // 如果机器人还没回复完return;}let input = document.getElementById("messageText");if (input.value === "") {return;}let body = {"prompt": input.value};ws.send(JSON.stringify(body));appendMessage(input.value, '用户')input.value = '';event.preventDefault();}document.getElementById("form").addEventListener('submit', sendMessage)ws.onmessage = function (event) {let body = JSON.parse(event.data);let status = body['status']if (status === 200) { // 如果回答结束了last_message_element = null;} else {history = body['history']last_message_element = appendMessage(body['response'], 'Qwen1.5-1.8B-chat', last_message_element)}};
</script>
</body>
</html>
- 运行结果
- 压测代码
locustwebsocket.py
import json
import timeimport websocket
from locust import User, TaskSet, task, events
import randomclass WebSocketClient(object):def __init__(self, host):self.host = hostself.ws = websocket.WebSocket()def connect(self, burl):start_time = time.time()try:self.conn = self.ws.connect(url=burl)except websocket.WebSocketTimeoutException as e:total_time = int((time.time() - start_time) * 1000)events.request_failure.fire(request_type="websocket", name='urllib', response_time=total_time, exception=e)else:total_time = int((time.time() - start_time) * 1000)events.request_success.fire(request_type="websocket", name='urllib', response_time=total_time, response_length=0)return self.conndef recv(self):return self.ws.recv()def send(self, msg):self.ws.send(msg)class WebsocketLocust(User):def __init__(self, *args, **kwargs):super(WebsocketLocust, self).__init__(*args, **kwargs)self.client = WebSocketClient(self.host)class SupperDianCan(TaskSet):@taskdef test(self):self.url = 'http://127.0.0.1:6006'self.data = {}self.client.connect(self.url)while True:recv = self.client.recv()print(recv)if eval(recv)['type'] == 'keepalive':self.client.send(recv)else:self.client.send(self.data)
class WebsocketUser(TaskSet):host = "http://127.0.0.1:6006"client = Nonedef on_start(self):self.client = WebSocketClient("ws://127.0.0.1:6006/ws")self.client.connect()@taskdef send_message(self):# 发送的订阅请求num = random.randint(0, 10)prompt = f"世界上第{num}高的山峰是什么"self.client.send(json.dumps({'prompt': prompt}))response = self.client.recv()print(json.loads(response))class WebsiteUser(User):tasks = [WebsocketUser]min_wait = 3000max_wait = 6000stop_timeout = 5
