Command to train on TPU for $(\mu,p_{drop})=(2,0.1)$. The path /pathOfTrainValMeta/  contains A) the folder "train" with all the sub-folder 
with the images for training ImageNet per class, B) the folder "val" with all the sub-folder with the images for validation on ImageNet. The
/output/Path is where the checkpoints will be saved.

!python /Path/To/launch_xla.py    --num-devices 8    /Path/To/train.py     /pathOfTrainValMeta/ \
    --experiment  Coef2_ProbDrop01  --output  /output/Path   --checkpoint-hist  1\
    --Coef-LossDisentangle 2.0   --dp-prob 0.1\
    -j 1 -b 192 --lr 0.6  --input-size 3 160 160 \
    --epochs 240 --warmup-epochs 5  --sched cosine --aa rand-m7-mstd0.5-inc1 --weight-decay 1e-4 --reprob 0.4 --recount 3 --remode pixel 




For google Colab we used the following installations:
!pip install git+https://github.com/rwightman/pytorch-image-models.git@bits_and_tpu
!pip uninstall -y torch
!pip install torch==1.8.2+cpu torchvision==0.9.2+cpu -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html
!pip install  torchtext==0.9.1 -f https://download.pytorch.org/whl/cu101/torch_stable.html
!pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.8-cp37-cp37m-linux_x86_64.whl
!pip install --quiet pytorch-lightning
!export XRT_TPU_CONFIG="localservice;0;localhost:51011"
