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.
- OS: Ubuntu 20.04
- CPU: Intel (R) Core (TM) i7-6950X CPU @ 3.00GHz
- GPU: Single NVIDIA TITAN X Pascal
- Pytorch 1.13.1
- Python 3.7.16
- nas-bench-201 1.3
- CUDA 12.0
-
Download the image datasets.
-
Place the folders in ~path_to_lcnas/datasets/
-
Download NAS-Bench-201.
-
Place the .pth file in ~path_to_lcnas/datasets/
./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.