FlowerFormer: Empowering Neural Architecture Encoding using a Flow-aware Graph Transformer

Official implementation of our paper, FlowerFormer: Empowering Neural Architecture Encoding using a Flow-aware Graph Transformer (CVPR 2024)

Prerequisites

• Python 3.10
• Pytorch 1.13.1
• Pytorch Geometric 2.2.0

Our model implemented on with GraphGPS framework as the backbone.

Please, install GraphGPS from link

Dataset

In our paper, we used 5 datasets: NAS-Bench-101, NAS-Bench-201, NAS-Bench-301, NAS-Bench-Graph, NAS-Bench-ASR.

We provided preprocessed datasets (PyG format) here.

Please place the data in ./data folder.

To run experiments

Run python experiments.py [config_path] with corresponding config path.