This is a pytorch implementation of the DARTS.
Since the offical released code does not support the pytorch 0.4 or later, here we reimplement the DARTS with pytorch 1.0.
nohup python3 train_search.py > train_search.log &
nohup python3 train.py > train_model.log &
We train the search model via a single Tesla V100 GPU for 4 days, while we train the model on cifar-10 for 2 days with two Telsa v100 GPUS.