Skip to content

Latest commit

 

History

History
38 lines (27 loc) · 1.45 KB

Docker.md

File metadata and controls

38 lines (27 loc) · 1.45 KB
Raw

Docker

You have to use the Docker to complete all your assignments on servers.

Instructions

  • We will assign an IP address and a port number for lucky ones. They can log in via ssh. e.g. ssh root@40.104.61.196 -p 8100
  • The initial password of docker is root. The first thing is to update your password when you log in successfully.
  • Run cd into your home directory.
  • Run git clone https://github.com/PeiqinSun/tf-tutorials.git get a repo for course.
  • Run cd tf-tutorials/01-svhn into your first homework.
  • Run CUDA_VISIBLE_DEVICES=${NUM} python train.py to start. NUM can be 0~7.
Warnings
  • You must use your real name and real id. All containers that do not conform to the naming convention will be cleared!!
  • Don't interrunpt the expriemnt during the data filling stage, otherwise you will generate a large file called core in your directory.

GPU Usage

When running your train script, you should use environment variable CUDA_VISIBLE_DEVICES to specify which GPU your program is running on.

CUDA_VISIBLE_DEVICES=0 python train.py

To monitor GPU usage, your can use

watch nvidia-smi

Tmux

If your program is still running, but you want to temporarily exit the terminal. You can use tmux, a terminal multiplexer software. If you want get more information about tmux, please access http://cenalulu.github.io/linux/tmux/. Run sudo apt install tmux to install tmux.