2021-07-07

TENSORFLOW里面的大坑：Loaded runtime CuDNN library: 7.4.1 but source was compiled with: 7.6.0. CuDNN library

这个错误后面跟着的错误是Faile 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.

解决办法：这是由于anaconda可能自身带有7.6.1的cudnn，所以理论上不安装cudnn也可以直接运行tensorflow（但是我没试过）。解决的方法是在cudnn官网上下载7.6.0的版本，重新把下载的文件夹里面的内容放到cuda对应的bin/lib/include文件夹里面去。问题得以解决。
cudnn下载：https://developer.nvidia.com/rdp/cudnn-archive。找对版本！