Cuthon is a simple Python script to avoid setting CUDA_VISIBLE_DEVICES when running python programs on a GPU. The script will find the first unused GPU(s) then run the program as usual. At its simplest:
cuthon my_program.py
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).
Install with pip:
pip install cuthon
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 --
The output will be:
usage: cuthon.py [-h] [-n NUM_GPUS] [-l]
Select the first unused GPU(s) and run Python. To pass the script arguments
specify '--' between cuthon arguments and arguments to be passed through to
your script. If '--' is not specified, then all arguments will be passed
through.
optional arguments:
-h, --help show this help message and exit
-n NUM_GPUS, --num_gpus NUM_GPUS
The number of GPUs to use.
-l, --least_used Switch from an unused to a least-used policy.
For example, to run on two available GPUs execute:
cuthon -n 2 -- train_model.py