EfficientNet-B3 training result

## Training 

To train a model, run scripts with the desired model architecture and the path to the ImageNet dataset:
# Link the dataset from source to target
ln -s {data/path} EfficientNet-B3/pycls/datasets/data

# 1p train 1p
# run 1 epoch
bash test/train_performance_1p.sh --data_path={data/path} # train performance

# 8p train 8p
# run 1 epcoh
bash test/train_performance_8p.sh --data_path={data/path} # train performance
# run 100 epoch, running time: 27h  
bash test/train_full_8p.sh --data_path={data/path} # train accuracy

# 1p eval 1p
bash test/train_eval_8p.sh --data_path={data/path}