The code for Joint Neural Architecture Search and Quantization

• pytorch 0.4.1
• python 3.6
• tensorboardX

python train_search.py --data 'path_to_dataset'

For CIFAR-10 models, we rely on the traing code from DARTS to train and evaluate the searched network.

For ImageNet models, we rely on the traing code from RENASNet to train and evaluate the searched network.

Our implementation about quantization highly replies on the quantized_distillation.

We also thanks for the inspiring work DARTS.