为什么Pytorch多卡训练容易导致GPU显存不释放

Others 2021-11-27 09:48:50 views: null

      使用PyTorch设置多线程（threads）进行数据读取（DataLoader），其实是假的多线程，他是开了N个子进程（PID都连着）进行模拟多线程工作，所以你的程序跑完或者中途kill掉主进程的话，子进程的GPU显存并不会被释放，需要手动一个一个kill才行

Guess you like

Origin blog.csdn.net/hxxjxw/article/details/121107486
Recommended
Ranking
How cascading public network servers and network companies with free softether, forming a network penetration
axios package
C implementation creates multiple txt file and named to the natural numbers, then the resulting decimal data written txt file
ctfhub技能树 信息泄露 目录遍历 PHPINFO 备份文件下载
The real version of graduation = #"Massey University graduation certificate in New Zealand" is exactly the same as the original MQU certificate
Examples of simple web application architecture
MATLAB graphics drawing-polar and logarithmic images
java里的this和super关键字使用（附代码实例操作）
Markdown을 사용하여 다양한 플랫폼과 호환되고 보기 쉽고 매우 아름다운 Readme 또는 기사를 작성하는 방법은 무엇입니까? 자세한 그림, 도구, 예제가 포함된 템플릿이 첨부되어 있습니다.
What does java startup hardcode correspond to?
Daily
More
2024-04-23(30)
2024-04-22(5)
2024-04-21(0)
2024-04-20(6)
2024-04-19(5)
2024-04-18(0)
2024-04-17(31)
2024-04-16(23)
2024-04-15(5)
2024-04-14(0)

Code World

© 2018 codetd.com