This code is based on the implementation of DARTS.
To install requirements:
pip install -r requirements.txt
To train the model architectures, run this command:
python main.py
To search the model architectures by MF-NAS, set the hyper-parameters in
config.py
as such:
Hyper-parameters | Value |
---|---|
dataset_root | path_to_dataset |
We provide one of the model architecture searched by MF-NAS and the trained weights on CIFAR10 of the model.
To evaluate the performance of pre-trained model on CIFAR10, run:
python evaluate.py
One of the model achieved by MF-NAS on CIFAR10:
Model name | Top 1 Acc | Params (M) |
---|---|---|
MF-NAS | 97.52% | 3.3 |