goodluck

Installation

pip install git+https://github.com/Rhyssiyan/goodluck.git

如果你想将语言设置恢复为英文

locale-gen 'en_US.UTF-8' update-locale LC_ALL='en_US.UTF-8'

Usage

1. Run a program on node satisfying some requirements

• 找到一个空闲的TitanV节点,在其上运行nvidia-smi goodluck run "nvidia-smi" --ngpu 1 --card 'v'

Common parameters:

• ngpu: how much gpu you want to use
• env: The environment you want to source(default is conda environment)
• virt_env: if true, source virtual environment
• card: gpu card type. Legal values:'all' | '1080' | 'm40' | 'xp' | 'v' | 'k40' | 'v100' | 'p40'
• exit: If program ends, exit from corresponding node

2. Run multiprogram according to some config

3. Observe what nodes are free in AI cluster

• See all free nodes: goodluck watch
• See free nodes with TITAN V: goodluck watch --card 'v'
• Watch all nodes except the nodes with k series card. watch -c -n 0.1 "goodluck watch --noicon --card 'all,-k' "

4. Observe what nodes are free in P40 Cluster goodluck p40_watch
5. Wrap your program with tmux

• Only wrap with tmux goodluck wrap
• Wrap your command goodluck wrap 'nvidia-smi' --env xxx

Caution:

• 不要在你的command中再指定显卡, i.e. CUDA_VISIBLE_DEVICES=xx