This repository contains code for two papers:
- TABv2: A Faster Ternary and Binary Neural Network Inference Library on the Edge
- Faster Ternary and Binary Neural Network Inference on CPU by Reducing Popcount Overhead
TABv2 is the journal version of FasterTNN with new quantization and bitwise GEMM algorithms on GPU and more experiments. Thus, the CPU part of TABv2 shares the code of FasterTNN. Please refer to the README.md inside each folder for detailed experiment setups.
File organization and main contributors:
- FasterTNN (TABv2_CPU): Olivier Fischer
- TABv2_GPU: Guanshujie Fu
If you find this repository useful, please cite the following paper(s):
@inproceedings{FasterTNN_ISLPED_2025,
title={Faster Ternary and Binary Neural Network Inference on CPU by Reducing Popcount Overhead},
author={Olivier Fischer and Shien Zhu and Gustavo Alonso},
booktitle={2025 IEEE/ACM International Symposium on Low Power Electronics and Design (ISLPED)},
pages={1--6},
year={2025},
organization={IEEE}
}