TensorFlow报错：This is probably because cuDNN failed to initialize

Peanut_范

于 2021-06-10 16:04:12 发布

阅读量392

点赞数

CC 4.0 BY-SA版权

分类专栏： Tensorflow的学习文章标签： tensorflow cuDNN

本文链接：https://blog.csdn.net/u013841196/article/details/117783176

Tensorflow的学习专栏收录该内容

22 篇文章

订阅专栏

本文介绍了解决TensorFlow中出现的cuDNN初始化失败的问题，此问题会导致卷积操作失败。通过调整GPU内存分配策略，可以有效避免该错误，确保训练过程正常进行。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

TensorFlow报错：This is probably because cuDNN failed to initialize

tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found.
  (0) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
         [[{{node efficientnet-b0/model/stem/conv2d/Conv2D}}]]
  (1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
         [[{{node efficientnet-b0/model/stem/conv2d/Conv2D}}]]
         [[bottleneck/_777]]
0 successful operations.
0 derived errors ignored.

解决方法：

config = tf.compat.v1.ConfigProto(gpu_options=tf.compat.v1.GPUOptions(allow_growth=True))
sess = tf.compat.v1.Session(config=config)

之前的代码：sess = tf.Session()