- Background situation: The training is normal and the verification is normal, but when using tensorboard to visualize the specific training situation, an error is suddenly reported
ImportError libcublas.so.9.0
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory
- Looking up the situation on the network, most of them say that the versions of tensorflow and cuda do not match, but considering that the training and testing processes are normal, I disagree
- The final consideration is that the dynamic link library is not normally linked, and the correct method is to add it in the environment variable or configuration file. The environment variables have been configured before, but they still don't work well.
-
- Just try to use the configuration file:
cat /etc/ld.so.conf
include /etc/ld.so.conf.d/*.conf
-
- Create a new configuration file:
vi /etc/ld.so.conf.d/cuda.conf
添加如下内容:
/usr/local/cuda-9.0/lib64
-
- Finally execute the following command:
ubuntu@ubuntu:/usr/local/cuda-9.0/lib64$ ldconfig -v | grep libcu
Then the problem is solved~tensorboard can be used normally.
Reference link:
TensorFlow ImportError libcublas.so.9.0 cannot be used in Crontab and sudo