我的最终配置:WIN10 anaconda cuda10.0 cudun7.4 python3.7 tensorflow1.13.1 numpy 1.16.4
遇到的问题一:
我初始安装的cudnn版本是 cudnn-10.0-windows10-x64-v7.6.5.32,运行程序时出现如下的问题:
Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize(错误提示中的一段)
解决办法:将cudnn版本从7.6.5降到7.4.1,运行程序后,没出现上述问题。
遇到的问题二:
FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
_np_qint8 = np.dtype([("qint8", np.int8, 1)])
出现问题的原因是numpy版本与tensorflow版本不兼容;
解决办法:将numpy版本从1.18.1降成1.16.4,问题解决。
参考自:
- https://github.com/tensorflow/tensorflow/issues/28326
- https://github.com/tensorflow/tensorflow/issues/31249