Content

  1. Links
  2. Installation
    1. Torch Related
    2. Venv
  3. System&Terminal
  4. Python
  5. Docker
  6. No Hang up
  7. Errors
    1. 一些文字没有显示
    2. dpkg: 错误: 另外一个进程已经为 dpkg frontend lock 加锁
  8. Git
    1. Git checkout and git switch/restore
    2. Merge one branch to another branch
    3. Modify the latest commit:
    4. Submodule Tracking
  9. CUDA
    1. Assign Process to GPU

Links

https://github.com/HuangJianxjtu/linux_learning/blob/master/ubuntu20_setup_notes.md

Installation

conda create -n myenv python=3.8 #create new env
conda remove -n myenv --all #delete the whole env

  • install torch/torchvision/torchaudio with specefic cudatoolkit:pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117

  • Install whl file: pip install ***.whl

  • 国内源 -i https://pypi.tuna.tsinghua.edu.cn/simple

Venv

  • python3 -m venv venv 创建新环境
  • source venv/bin/activate 激活环境
  • deactivate 退出环境

System&Terminal

  • Supervise the GPU usage watch -n 10 nvidia-smi

  • Look for proxy env|grep -i proxy

  • Check the system archtecture uname -m

  • Package Management:

    • Installationsudo dpkg -i package_name.deb
    • Uninstallation sudo dpkg -r package_name

  • Show the path of the Python intepreter:

    import sys
path = sys.executable
print(path)
    • Select the Python Intepreter in VSCode:
      • Ctrl+Shift+P to open the control panel
      • Choose Python: Select Interpreter

  • Vim Handbook

Python

python -c "print('Hello from command line')" #execute python code directly from CLI

python -c "import torch; print(torch.cuda.is_available()); print(torch.__version__)"

python -m pip install package_name #install packages using pip

python [main.py](http://main.py) -h get the subsequent commands

Docker

docker images 查看当前所有docker镜像

docker rmi [REPOSITORY:TAG] 删除指定镜像

No Hang up

以下命令在后台执行 root 目录下的 runoob.sh 脚本（注意&才能让它在后台运行）
nohup /root/runoob.sh &

并且在当前目录下生成nohup.out文件

以下命令找到这个文件的PID
ps -aux | grep "runoob.sh"
进行删除
kill -9 进程号PID

Errors

一些文字没有显示

重启GNOME Shell

按下 Alt + F2
在弹出的运行对话框中输入字母 r
按下回车键

dpkg: 错误: 另外一个进程已经为 dpkg frontend lock 加锁

ps -e | grep apt 查看apt所有进程

sudo kill [PID] kill it

Git

Git checkout and git switch/restore

Actually same while the latter is more clear

# 切换到一个已存在的分支
git checkout develop
# 创建一个新分支并立即切换过去
git checkout -b new-feature
# 切换到一个特定的提交（会进入“分离头指针”状态）
git checkout <commit-hash>

# 撤销工作区中 a.txt 文件的修改，恢复到和暂存区一致的状态
git checkout -- a.txt
# 丢弃工作区和暂存区的所有修改，将 a.txt 文件恢复到上一次提交时的状态
git checkout HEAD -- a.txt
# 切换到一个已存在的分支 (和 checkout 作用相同)
git switch develop
# 创建一个新分支并立即切换过去 (用 -c 代替 -b，c 代表 create)
git switch -c new-feature
# 切换回上一个分支 (这是一个很方便的快捷操作)
git switch -

# 撤销工作区中 a.txt 的修改 (从暂存区恢复)
git restore a.txt
# 将暂存区的文件放回工作区 (unstage)
git restore --staged a.txt
# 同时撤销工作区和暂存区的修改，从上一次提交恢复文件
git restore --source=HEAD a.txt

Merge one branch to another branch

  • git pull --rebase means fetching all remote commits, and the changes will be after the newer commits
  • git push (suppose on main branch)
  • git checkout dev switch to another branch dev
  • git merge main merge main to dev

Modify the latest commit:

  • First stash the new changes, then usegit commit --amend ，use :wq to quit the Vim

Submodule Tracking

Sometimes, the changes in the submodule may not be tracked.

  1. Remove the original tracking: git rm --cached [File]

  2. Delete the .git file in submodule and delete the relative file in the root ..git if exists

  3. git add the file, same as other changes

CUDA

Assign Process to GPU

  • In Terminal: CUDA_VISIBLE_DEVICES=? python [main.py](http://main.py/) (GPU start from 0)
  • In Python: os.environ[“CUDA_VISIBLE_DEVICES”] = “?”
  • In System: export CUDA_VISIBLE_DEVICES=?
# cuda是否可用，可用返回true, 不可用返回false
torch.cuda.is_available()
# 查看gpu可用的数量
torch.cuda.device_count()
# 返回gpu名称, 默认从0开始
torch.cuda.get_device_name(0) # 如rtx 3090
# 返回当前gpu的索引
torch.cuda.current_device()