JAQ: Joint Efficient Architecture Design and Low-Bit Quantization with Hardware-Software Co-Exploration
Mingzi Wang, Yuan Meng, Chen Tang, Weixiang Zhang, Yijian Qin, Yang Yao, Yingxin Li, Tongtong Feng, Xin Wang, Xun Guan, Zhi Wang, Wenwu Zhu · 2025
Read JAQ: Joint Efficient Architecture Design and Low-Bit Quantization with Hardware-Software Co-Exploration by Mingzi Wang, Yuan Meng, Chen Tang, Weixiang Zhang, Yijian Qin, Yang Yao, Yingxin Li, Tongtong Feng, Xin Wang, Xun Guan, Zhi Wang, Wenwu Zhu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-05339.txt.
Description
Mingzi Wang and eleven co-authors build, in 2025 work, a frame that designs a neural net, its bit depth, and its chip in one sweep. Edge gear needs all three tuned at once, yet the joint space is vast and low-bit drills eat memory. Their frame trims both costs.
On the code side, a sparse drill path quantizes just the most touchy parts, which holds memory down. On the chip side, a build net codes all tile plans at once, which cuts each search round to a blink. On a big image set the frame beat past work by near seven points of top-call truth. Chip search time fell just as hard.
Catalog record
| CREATOR | Mingzi Wang, Yuan Meng, Chen Tang, Weixiang Zhang, Yijian Qin, Yang Yao, Yingxin Li, Tongtong Feng, Xin Wang, Xun Guan, Zhi Wang, Wenwu Zhu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-05339