cnn-models
cnn-models copied to clipboard
cvjena
train on multi gpus
Hi,
I tried to finetune with your model. It works well on single gpu but not on multi gpus. Is it a inhere limitation or there is something I need to work around? Thanks.
Hi!
Which model are you using? Multi GPU shouldn't make a difference for VGG or AlexNet, but can make a difference for ResNet. For ResNet, please make sure that each GPU gets at least 8, better 16, images of each batch.
Another thing to take care of is the learning rate. It should be scale linearly with the batch size, which you probably scale when using multiple GPUs.
Am 19. Februar 2019 18:47:07 schrieb Li Yu [email protected]:
Hi, I tried to finetune with your model. It works well on single gpu but not on multi gpus. Is it a inhere limitation or there is something I need to work around? Thanks. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.
Thank you for your response. I finally found that there is something wrong with the environment. I tried to run it on another server and it worked like a charm.