search

huaaaliu / RGBX_Semantic_Segmentation

MIT License
315 stars 29 forks source link

training with a single GPU #12

Closed jaeehooon closed 2 years ago

jaeehooon commented 2 years ago

Hello. I appreciate for me to making this code available!

I would like to ask a question about using 'a single GPU for training'. Is this command only for training with a single one??

python train.py -d 0 ??

In this command that you proposed, I don't know how to train with a single one. $ CUDA_VISIBLE_DEVICES="GPU IDs" python -m torch.distributed.launch --nproc_per_node="GPU numbers you want to use" train.py

Is it correct with this one? $ CUDA_VISIBLE_DEVICES="0" python -m torch.distributed.launch --nproc_per_node="1" train.py ??

Even CUDA_VISIBLE_DEVICES is not working on Ubuntu!

Thank you. Best Regards

jamycheung commented 2 years ago

Hi, thanks for your interest. You are right, the command can be used if you train with a single GPU. For example, CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --nproc_per_node=1 train.py should work fine.

1600337881 commented 2 years ago

Hi, thanks for your interest. You are right, the command can be used if you train with a single GPU. For example, CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --nproc_per_node=1 train.py should work fine.

Hi,Is this setting in engine.py QQ图片20221116164303 correct?