Install CUDA and cuDNN before installing tensorflow-gpu. For specific installation steps, please refer to the previous article: Record the process of correctly installing CUDA and cuDNN on Win10 (record some pitfalls)
The CUDA version installed on my computer is 10.2, and the cuDNN version is 7.6.5. According to the comparison table on the TensorFlow official website:
tensorflow-gpu 2.1.0, 2.2.0, 2.3 . 0 can be installed, it is recommended to install the lower version 2.1.0 here, but I installed 2.3.0.
Enter Anaconda Prompt, enter:
pip install tensorflow-gpu==2.3.0 -i https://pypi.tuna.tsinghua.edu.cn/simple
The installation was successful!
Go to GitHub to view the version correspondence between tensorflow and tensorflow_federated:
Install version 0.17.0 here:
pip install tensorflow_federated==0.17.0 -i https://pypi.tuna.tsinghua.edu.cn/simple
It is worth noting that In the process of installing tensorflow_federated, the cpu version of tensorflow will be automatically installed, and the version of tensorflow automatically installed when tensorflow_federated 0.17.0 is installed is 2.3.4, then in the environment at this time The CPU version and GPU version of tensorflow exist at the same time, as shown below:
If Both the CPU version and the GPU version are installed in the environment, and the GPU version runs by default.