alrao
alrao copied to clipboard
memory consumption too high
Hi First of all, I really liked the paper and the implementation. but when i ran it for VGG19 network, it took almost 10.5GB of GPU RAM( I have 1 1080Ti 11GB) Is it normal to have this much memory consumption?
Hi,
I have tested the following setup (with Alrao):
python main_cnn.py --epochs=1000 --early_stopping --use_alrao --minLR=-5 --maxLR=1 --n_last_layers=7 --data_augm --data_path=data_cifar --model_name=VGG19
Result: 2119MiB memory usage.
Without Alrao (but with our script):
python main_cnn.py --epochs=1000 --early_stopping --data_augm --data_path=data_cifar --model_name=VGG19
Result: 2027MiB memory usage.
(be careful, I changed the file main_cnn.py
)
With another (non-optimal) script: 1381MiB memory usage.
What script did you run? With which options?