train_fb_poseidon

#!/bin/bash
docker run -it \
--shm-size 100GB \
-u 0:0 \
--gpus all \
--name="csj-runcmd-$(date +%s)" \
-v $(pwd):/app/ \
-v $(pwd)/imagenet:/app/imagenet \
csi \
python3 -m torch.distributed.launch --nproc_per_node=8  \
./pytorch-image-models/train.py  \
	./imagenet/              \
	--val-split 'val'       \
	--model fbnetv3grayworld_b \
	-b 384 				    \
	--sched step 		    \
	--epochs 350            \
	--decay-epochs 2.4      \
	--decay-rate .97        \
	--opt rmsproptf 		\
	--opt-eps .001 	        \
	--warmup-lr 1e-6 		\
	--weight-decay 1e-5     \
	--drop 0.2              \
	--drop-connect 0.2      \
	--model-ema             \
	--model-ema-decay 0.9999 \
	--aa rand-m9-mstd0.5    \
	--remode pixel 			\
	--reprob 0.2            \
	--amp 					\
	--lr .048 				\
	-j 12                   \
	--output ./out/fbnetv3grayworld_b | tee out.txt