Ubuntu 20.04 CUDA+cuDNN+tensorflow+Bazel

It's for my own record.

Environment: Laptop (Ubuntu 20.04) 

                      NVIDIA  [GeForce GTX 1050 Ti Mobile]

                              

In order to install the Navidia environment to improve the speed of YOLOv5-lite, the following packages should be installed firstly.

Check the version of Python3 first.

The versions of these packages should match as follows(first line). 

Tensorflow

https://www.tensorflow.org/install/source#ubuntuhttps://www.tensorflow.org/install/source#ubuntu

If you encounter "Error - launchpadlib requires "testresources"  which is not installed", try the following code

sudo apt install python3-testresources

If the version of protobuf is too high, try to reinstall it,

sudo apt remove protobuf

then use "protoc --version" to check the version.

protoc --version

If there is no response, you can reinstall the protobuf by

sudo apt install snapd
sudo snap install protobuf --classic

Then the version of protobuf is supposed to be 3.14.0

CUDA+cuDNN

Ubuntu 20.04安装CUDA & CUDNN 手把手带你撸_哈希Map的博客-CSDN博客_ubuntu20.04安装cuda新手先看这之前一直使用CPU做训练，最近手上多了台单卡1080Ti显卡主机，于是开始研究GPU训练。用GPU训练一定会使用CUDA了，刚开始接触CUDA非常非常头痛，对小白很不友好，我几乎整了整整一天，中途想放弃过很多次，最后终于弄好了，回头看看其实不是很难，但是网上文章又多又杂，确实让我来回鼓捣了很久，于是写下这篇文章，希望对新入门GPU机去学习，想安装CUDA的朋友们一点帮助，少走弯路，不要像我一样浪费好多时间。新手最难理解的就是CUDA和cuDNN是啥了，以及安装过程，步骤。 这里先简单说一下https://blog.csdn.net/h3c4lenovo/article/details/119003405However, you'd better use Section 5 in the following link to change the PATH

Ubuntu 20.04安装CUDA 11.4, cuDNN 8.24 - 知乎温馨提示：如果你装CUDA是为了安装GPU版本的pytorch，那么其实不用看我下面关于CUDA安装的内容，因为你使用pytorch官网的安装命令，它会自动帮你安装相应的CUDA，但是cuDNN可能还得你手动安装。如果是安装TensorFl…https://zhuanlan.zhihu.com/p/424817205

Bazel

Installing Bazel on Ubuntu - Bazel 3.7.0https://docs.bazel.build/versions/3.7.0/install-ubuntu.html

You should always be cautious with the versions of these packages.