Single GPU training

Question

Single GPU training

alfredgu001324 opened this issue a year ago · 1 comments

Hi, thank for the great work.

I am trying to reproduce your result using just one single GPU. But the following error occurs:
RuntimeError: Default process group has not been initialized, please make sure to call init_process_group.

by running this command:
python tools/train.py /home/guxunjia/project/StreamMapNet/plugin/configs/nusc_baseline_480_60x30_30e.py --deterministic --no-validate

Answer 1 · 2024-01-10T08:05:57.000Z

You should start the training by torch.distributed.launch even if you only use one gpu. What's more, according to the comment code of this repo, the bevformer backbone only support distributed training.