If you use ANT in your research, please cite our paper:
@inproceedings{guo2022ant,
title={ANT: Exploiting Adaptive Numerical Data Type for Low-bit Deep Neural Network Quantization},
author={Guo, Cong and Zhang, Chen and Leng, Jingwen and Liu, Zihan and Yang, Fan and Liu, Yunxin and Guo, Minyi and Zhu, Yuhao},
booktitle={2022 55th IEEE/ACM International Symposium on Microarchitecture (MICRO)},
pages={1414--1433},
year={2022},
organization={IEEE}
}
This repository contains the source code for reproducing the experiments in the paper "ANT: Exploiting Adaptive Numerical Data Type for Low-bit Deep Neural Network Quantization" at MICRO'22.
ant_quantization contains the ANT framework with PyTorch.
ant_simulator contains the performance and energy evaluation of ANT.
├── ant_quantization # The ANT framework with PyTorch.
│ ├── antquant # Quantization framework of ANT.
│ ├── BERT
│ │ ├── download_glue_data.py # Download for GLUE data.
│ │ └── scripts # Reproduce the experiment data in Figure 12.
│ ├── ImageNet
│ │ └── scripts # Download checkpoints and reproduce the experimental data in Figure 12.
│ ├── quant # Quantization CUDA kernel.
│ └── result # Our test results.
├── ant_simulator # The performance and energy evaluation of ANT.
│ ├── results # Our test results.
│ └── run_ant.py # The scripts are for reproducing the experiment data in Figure 13.
Licensed under an Apache-2.0 license.