Latency-Constrained Neural Architecture Search Method for Efficient Model Deployment on RISC-V Devices

This repository contains code and RISC-V latency dataset of the paper.

Environment

Hardware

• OS: Ubuntu 20.04
• CPU: Intel (R) Core (TM) i7-6950X CPU @ 3.00GHz
• GPU: Single NVIDIA TITAN X Pascal

Software

• Pytorch 1.13.1
• Python 3.7.16
• nas-bench-201 1.3
• CUDA 12.0

Setup

Image Datasets

1. Download the image datasets.

2. Place the folders in ~path_to_lcnas/datasets/

NAS-Bench-201

1. Download NAS-Bench-201.

2. Place the .pth file in ~path_to_lcnas/datasets/

Running LC-NAS

./run.sh

This command uses latency predictor to search architecture within latency constraint 500 ms in different sample numbers on CIFAR-10. The resluts will be shown in the cmd.