服务化部署PaddleOCR的QPS过低
#13841
Replies: 2 comments
-
|
尝试循环跑100遍测试一个平均值看看 |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
应该是直接用了pipline_http_client.py的原因，后面改成用jmeter来做接口测试，QPS就有15+了。感谢回复！
…On Sep 11, 2024 at 16:23 +0800, WangZ ***@***.***>, wrote:
尝试循环跑100遍测试一个平均值看看
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
UserWangZz
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
按照官方教程服务化部署PaddleOCR-py，最终测试发现QPS只有4-5，距离官方指标60还有很大差距。想请教下还有没有优化的空间。下图是pipeline.tracer的截图，相较于其他指标，in的数值异常高，请问是什么原因？
Beta Was this translation helpful? Give feedback.
All reactions