search

facebookresearch / moco

PyTorch implementation of MoCo: https://arxiv.org/abs/1911.05722
MIT License
4.83k stars 794 forks source link

Multi-node #97

Open techmonsterwang opened 3 years ago

techmonsterwang commented 3 years ago

Thanks for the great repo. I notice that the readme file only give a suggestion on training on one node with 8 gpus. So I wonder how does the code support for multi-nodes, such as:

For node 0

python main_moco.py \ -a resnet50 \ --lr 0.03 \ --batch-size 256 \ --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 2 --rank 0 \ [your imagenet-folder with train and val folders]

For node 1

python main_moco.py \ -a resnet50 \ --lr 0.03 \ --batch-size 256 \ --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 2 --rank 1 \ [your imagenet-folder with train and val folders]

wenyuqing commented 3 years ago

Hi, have you successfully implemented multi-node version of moco? I meet some problem when running with multi-node. The buffer of queue seems not synchronized and the variable ptr is not 0 at the beginning.