Solve the Linux Tensorflow2.0 installation problem

conda update conda
pip install tf-nightly-gpu-2.0-preview
conda install https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/cudnn-7.3.1-cuda10.0_0.tar.bz2
conda install https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/cudatoolkit-10.0.130-0.tar.bz2

illustrate:

• First you need to update conda
• The latest version of tf2.0 is installed
• cudnn7.3.1 and cudatoolkit-10.0 versions can be downloaded and installed locally
  

wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/cudnn-7.3.1-cuda10.0_0.tar.bz2
conda install cudnn-7.3.1-cuda10.0_0.tar.bz2

Errors and solutions

Old library problem

ERROR: Cannot uninstall 'wrapt'. It is a distutils installed project and thus we cannot accurately determine which files belong to it which would lead to only a partial uninstall.

The old version has many dependencies and cannot be deleted clearly. In this case, you should ignore the old version upgrade, which is the following solution: pip install tf-nightly-gpu-2.0-preview --ignore-installed wrapt

NumPy version problem

Another problem is that there is an old version of numpy. You can use pip to uninstall numpy until it prompts that there is nothing to uninstall, and then reinstall numpy

Driver issues

tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version

This is because the driver version does not match. You can download the cuda10.0 driver (which is the same as above) from the NVIDIA official website.

Installation command: https://juejin.im/post/5cce44e3f265da036902a89c , then confirm all the way, and finally use watch nvidia-smi


View the results:

Testing and other

Tests available:

import tensorflow as tf
print(tf.__version__)
print(tf.keras.__version__)

if tf.test.is_gpu_available():
  device = "/gpu:0"
else:
  device = "/cpu:0"

print(device)

Reduce tensorflow output information

TensorFlow has four levels of log information, in ascending order of importance: INFO (notification) < WARNING (warning) < ERROR (error) < FATAL (fatal)

tf.compat.v1.logging.set_verbosity('ERROR')

or

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

Tensorflow2.0 prompts problems under pycharm

Tensorflow2.0 usually uses keras through tensorflow.keras, but pycharm does not prompt it. The reason is that the actual keras path is placed in tensorflow/python/keras, but there is no python directory in tensorflow in the program. The solution is as follows:

try:
  import tensorflow.python.keras as keras
except:
  import tensorflow.keras as keras

In this way, Pycharm can have prompts, and there is no need to modify the code when the program is running.

Summarize

The above is the editor's introduction to solving the Linux Tensorflow2.0 installation problem. I hope it will be helpful to everyone. If you have any questions, please leave me a message and I will reply to you in time. I would also like to thank everyone for their support of the 123WORDPRESS.COM website!
If you find this article helpful, please feel free to reprint it and please indicate the source. Thank you!