Skip to content

Running programs on multiple GPUs

Jump to bottom
Cryolite edited this page Jan 17, 2023 · 3 revisions

The learning programs in this project support training with multiple GPUs. For this purpose, torchrun is used. If the command to run a learning program using a single GPU is as follows

$ docker run ..... image-name python3 -m path.to.module OPTIONS...

then the corresponding command to use multiple GPUs becomes, for example,

$ docker run ..... image-name torchrun --standalone --nproc_per_node gpu -m path.to.module OPTIONS...

For more information on how to use torchrun, see torchrun --help and others.

Clone this wiki locally