1. The first:
I have the number of virtual environment:

Tensorflow2GPU is tf2.0.0 GPU version, tf23GPU tf2.3.0 version, including: tf23GPU is cloned by tensorflow2GPU environment to come over and then upgrade version of the TF,
The two virtual environment in shell terminal can be normal use, the TF also can each normal import: as follows:




2 a description of the problem:
. Then use the jupyter, in TF2.3 versions of virtual environment import tensorflow got 2.0.0 version of is very strange, and the next word import is normal 2.0.0 version,
Arouse jupyter script is as follows:


In tf23GPU environment (TensorFlow version 2.3.0 GPU), import the tf library is 2.0.0 version.


In tensorflow2GPU environment (TensorFlow version 2.0.0 GPU), import the tf library version is 2.0.0.



3. My graphics card information: [because is installed before tf2.0.0 version, so drive is subject to a 2.0.0 but I after the cloning of a virtual environment, the tf2.3.0 in virtual environment for the installation of version 10.1, cudatoolkit cudnn 7.6.5 version, accord with the requirement of 2.3.0]


4. My guess:
I baidu online blogs and information, which may be associated with system environment variables of CUDA version, but I'm not sure whether my version information is as follows: [system only a 10.0 version]


5. Can you tell me what's the problem? How should solve? Seek the solution,

CodePudding user response:

, it seems, you this jupyter notebook should be adapted to tf2.0 version, according to the principle that, should be in each virtual environment a jupyter notebook can,
This problem and cuda, cudnn version has nothing to do, tf2.0 2.3 suitable are cuda10.1, cudnn7.6.5, so the upgrade to version 2.3 will not appear not adapt,
Advice, to 2.3 virtual environment to include a jupyter notebook, running, switch to the 2.3 virtual environment, run directly jupyter notebook,

CodePudding user response:

reference 1st floor Wang Dingbang response:

it seems that you this jupyter notebook should be adapted to tf2.0 version, according to the principle that, should be in each virtual environment a jupyter notebook can,
This problem and cuda, cudnn version has nothing to do, tf2.0 2.3 suitable are cuda10.1, cudnn7.6.5, so the upgrade to version 2.3 will not appear not adapt,
Advice, to 2.3 virtual environment to include a jupyter notebook, runtime, switch to the 2.3 virtual environment, run directly jupyter notebook,

Eldest brother, I has been solved before, I know what the problem is caused by the version of the version of the cuda, tf2.3 cuda10.1 support, but it does not support, tf2.0.0 tf2.0.0 cuda2.0.0 support, so I use a 2.3 I put cuda10.1 environment variables to cuda10.0 version in the front of the can, if I want to use tf2.0.0, I'll move cuda10.0 version of the environment variable to cuda10.1 version in the front of the can, I have already tested, indeed, tensorflow version (note: here are the gpu version)
Said of virtual environment business online, virtual environment simply provides an environment, call or to start the CUDA GPU resources, so the version is the key.