In the parent directory, run:
time python -m multiproc.py dawn_mod.py /home/ubuntu/data/cifar10 \
--save-dir /home/ubuntu/data/cf_train_save/wrn_v2_submission \
-a wrn_22 \
--epochs 1 --cycle-len 40 \
--fp16 --loss-scale 512 \
-b 128 \
--wd 2e-4 --lr 2.0 \
--use-clr 50,12.5,0.95,0.85 \
-j 8
on an Amazon p3.16xlarge instance. You will need to symlink (ln -s YOUR_FASTAI_DIRECTORY fastai) your fastai install inside the cifar directory for the python imports (or do git clone https://github.com/fastai/fastai.git inside the parent directory). Uses fastai master (current SHA1: https://github.com/fastai/fastai/commit/f9c3d22cdb79e9438c308e3baeb91f04944081fe) and pytorch master (.4) (https://github.com/pytorch/pytorch/commit/354dac976905ba718ea94a73c11587438ae86cde).
Dataset from:
http://files.fast.ai/data/cifar10.tgz
In the directory on paperspace using a volta (same setup as above), run:
python dawn_mod.py ~/data/cifar10/ --save-dir ~/data/cf_train_save/wrn_v5 -a wrn_22 --fp16 --loss-scale 512 --epochs 1 --cycle-len 30 --lr 1.5 --wd 1e-4 --use-clr 20,20,0.95,0.85