tensorflow引起的内存泄露

骑猪看日落 2021-11-22 13:24 337阅读 0赞

计算机配置内存12G，显存4G，运行有10分钟左右就提醒说内存不够用然后退出运行，直觉是程序bug问题   
终于找到靠谱解决方案：   
[http://cherishlc.iteye.com/blog/2324796][http_cherishlc.iteye.com_blog_2324796]  
这个博客的第二个给了大致的讲解，在给出的附录中：   
[https://stackoverflow.com/documentation/tensorflow/3883/how-to-debug-a-memory-leak-in-tensorflow/13426/use-graph-finalize-to-catch-nodes-being-added-to-the-graph\#t=201707221208374841351][https_stackoverflow.com_documentation_tensorflow_3883_how-to-debug-a-memory-leak-in-tensorflow_13426_use-graph-finalize-to-catch-nodes-being-added-to-the-graph_t_201707221208374841351]  
讲解更详细，并给出了解决的方法   
即：sess.graph.finalize() 使得整个graph变为只读的，不能再向图中添加任何节点   
我也明白了我的错误原因，在for循环中用了tf.convert\_to\_tensor方法，相当于不停地向图中添加节点，谢谢给出了解决方案的大牛们   
另一个附录：[https://github.com/tensorflow/tensorflow/issues/4151][https_github.com_tensorflow_tensorflow_issues_4151]  
通过一个示例给出了可行的替换方法，非常具有可行性

知乎的一个方法：[https://www.zhihu.com/question/58577743][https_www.zhihu.com_question_58577743]  
\---------------------   
作者：小妖精Fsky   
来源：CSDN   
原文：https://blog.csdn.net/appleml/article/details/75805797   
版权声明：本文为博主原创文章，转载请附上博文链接！

[http_cherishlc.iteye.com_blog_2324796]: http://cherishlc.iteye.com/blog/2324796
[https_stackoverflow.com_documentation_tensorflow_3883_how-to-debug-a-memory-leak-in-tensorflow_13426_use-graph-finalize-to-catch-nodes-being-added-to-the-graph_t_201707221208374841351]: https://stackoverflow.com/documentation/tensorflow/3883/how-to-debug-a-memory-leak-in-tensorflow/13426/use-graph-finalize-to-catch-nodes-being-added-to-the-graph#t=201707221208374841351
[https_github.com_tensorflow_tensorflow_issues_4151]: https://github.com/tensorflow/tensorflow/issues/4151
[https_www.zhihu.com_question_58577743]: https://www.zhihu.com/question/58577743