Select gpu ID

[yaoliUoA]

❓ Questions and Help

Hi

How to select gpu IDs to run experiments? I have multiple experiments to run on different gpus. It is not clear how to select gpu ID in the config file and there is no documentation about this. I think it is better to
add a "gpu id" option in config file.

[fmassa]

Hi,

The recommended way of selecting the GPU ids is to use the environment variable CUDA_VISIBLE_DEVICES.

So you'd run your experiments as follows:

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 tools/train_net.py --config-file configs/e2e_faster_rcnn_R_50_FPN_1x.yaml
CUDA_VISIBLE_DEVICES=2,3 python -m torch.distributed.launch --nproc_per_node=2 tools/train_net.py --config-file configs/e2e_faster_rcnn_R_50_FPN_1x.yaml
CUDA_VISIBLE_DEVICES=4,5,6,7 python -m torch.distributed.launch --nproc_per_node=4 tools/train_net.py --config-file configs/e2e_faster_rcnn_R_50_FPN_1x.yaml
CUDA_VISIBLE_DEVICES=0 python tools/train_net.py --config-file configs/e2e_faster_rcnn_R_50_FPN_1x.yaml

Because of that, I believe it's better not to add any option in the cfg to select the GPU ids.

But it would be a good idea to have a note in the README explaining this. Would you be willing to send a PR?

Given that the solution I mentioned is the best one I believe, I'm closing this issue. Please let me know if you have further questions.