博客

1.pytorch出现cuDNN error:CUDNN_STATUS_EXECUTION_FAILED

参考：

https://blog.csdn.net/qq_18483627/article/details/105878320?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-2.control&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-2.control

2.cuda9.0问题：failed to run cuBLAS routine cublasSgemm_v2: CUBLAS_STATUS_EXECUTION_FAILED

参考：

https://blog.csdn.net/ykf173/article/details/105542936/

https://www.freesion.com/article/7880800710/

3.pytorch出现RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED

解决：我是重新换了pytorch的版本

或者升级cuda，参考：

https://blog.csdn.net/OpenSceneGraph/article/details/104225951

4.pip版本问题You are using pip version 9.0.1, however version 20.2.3 is available

直接：python -m pip install -U pip

5.keras问题：original_keras_version = f.attrs[‘keras_version‘].decode(‘utf8‘)

AttributeError: ‘str’ object has no attribute ‘decode’

参考：

https://blog.csdn.net/qq_41251963/article/details/110132101

6.服务器中安装双cuda，我的是cuda10.1和cuda9.0

可能会遇到的问题：

gcc版本太新会导致cuda装不了。。。。降版本。。。

参考：

https://zhuanlan.zhihu.com/p/59123983

https://blog.csdn.net/yinxingtianxia/article/details/80462892

7.ubuntu16.04，cuda9.0  显卡2080ti快速配置深度学习环境：tensorflow1.8和pytorch，keras等等包。以下步骤，速度飞起。

1：conda create -n tf1.8 python=3.6

2：conda activate tf1.8

3：pip install keras==2.1.5 tensorflow-gpu==1.8 -i  https://pypi.doubanio.com/simple/

4：pip3 install Pillow -i  https://pypi.doubanio.com/simple/

5：pip install easydict opencv-contrib-python==4.0.0.21 Cython h5py lmdb mahotas pandas requests bs4 matplotlib lxml -i  https://pypi.doubanio.com/simple/