python测试大模型Qwen性能--待更新-先贴代码

from locust import HttpUser,task,between
import  time
import json


class QwenTPUTUser(HttpUser):
    host='http://201.201.201.35:9997/v1'
    wait_time=between(0.1,0.5)

    @task
    def test_tput(self):
        start_time=time.time()
        body={
          "model": "qwen2.5-instruct",
          "messages": [
            {
            "role": "system",
            "content": "You are a helpful assistant."
            },
            {
            "role": "user",
            "content": "我很开心，陪我聊聊天吧，今天吃几个鸡蛋"
            }
                  ]
            }
        headers={'Content-Type': 'application/json'}

        response=self.client.post(
                '/chat/completions',
                data=json.dumps(body),
                headers=headers)

        if response.status_code==200:
            resp_data=response.json(response)
            end_time=time.time()
            duration=end_time-start_time
            total_tokens=resp_data['usage']['total_tokens']
            TPUT=total_tokens/duration
            self.environment.events.request.fire(
                request_type="POST",
                name="TPUT",
                response_time=duration * 1000,  # 转为毫秒
                response_length=total_tokens,  # 用 Token 数替代响应大小
                context={"tput": TPUT}  # 额外存储 TPUT
            )