search

arthurdouillard / CVPR2021_PLOP

Official code of CVPR 2021's PLOP: Learning without Forgetting for Continual Semantic Segmentation
https://arxiv.org/abs/2011.11390
MIT License
145 stars 23 forks source link

don't use apex #41

Open yesokyin opened 1 year ago

yesokyin commented 1 year ago

Describe the bug A clear and concise description of what the bug is. Hello,because my server is 3090 and does not support CUDA10.2, how can I change the code if I don't use APEX? To Reproduce

Dataset: ... Setting: ... Command used or script used: ...

Expected behavior A clear and concise description of what you expected to happen.

Additional context Add any other context about the problem here.

QLYYLQ commented 6 days ago

just install another torch version is ok, i thought.

python -c "import torch; print(torch.__version__); print(torch.version.cuda)"

install pytorch with same cuda version for nvcc -V

Then you can go to the apex github to install it.

also you can follow this: https://discuss.pytorch.org/t/how-to-replace-apex-amp-by-pytorch-amp/182087, and use torch.nn.parallel.DistributedDataParallel to instead of apex's DDP.

Wish have a good experiment