Skip to content
/ cuthon Public

A simple tool to select the first free GPU(s) and run Python.

License

Notifications You must be signed in to change notification settings

awni/cuthon

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

cuthon

Cuthon is a simple Python script to avoid setting CUDA_VISIBLE_DEVICES when running python programs on a GPU. This tool is intended for a fairly niche use-case: running python programs on an interactive node which has more than one GPU. For those that have done this often you may be relieved at never having to run nvidia-smi followed by setting CUDA_VISIBLE_DEVICES again (when running a python program that is). This script should save you about 11.5 seconds per python program launch -- approximately the time it takes to run nvidia-smi followed by setting the CUDA_VISIBLE_DEVICES environment variable. I hope you enjoy those seconds back as much as I have.

Install

pip install cuthon

Usage

In general, use cuthon just like you would use python.

  • cuthon to launch a python repl.
  • cuthon -V to see the python version number.
  • cuthon train_model.py to run your program.

For help on available cuthon options type:

cuthon -h --

Note the -- separator. This is to distinguish between cuthon arguments and arguments to python. If the -- is present then all arguments before it are passed to cuthon and all arguments after are passed to python. If -- is absent all arguments are passed directly to python.

For example, to run on two available GPUs execute

cuthon --num_gpus=2 -- train_model.py

About

A simple tool to select the first free GPU(s) and run Python.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published