ScaleBITS: Scalable Bitwidth Search for Hardware-Aligned Mixed-Precision LLMs
Xinlin Li, Timothy Chou, Josh Fromm, Zichang Liu, Yunjie Pan, Christina Fragouli · 2026
Read ScaleBITS: Scalable Bitwidth Search for Hardware-Aligned Mixed-Precision LLMs by Xinlin Li, Timothy Chou, Josh Fromm, Zichang Liu, Yunjie Pan, Christina Fragouli (2026) 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-2602-17698.txt.
Description
Xinlin Li and team propose ScaleBITS, a mixed-precision bit-cutting frame for large language models. Pushing average bit width below four bits is hard because weight response varies sharply and principled bit sharing is lacking. Past fixes use ragged fine-grained mixed precision with high run-time overhead, or lean on rules of thumb and tightly constrained sharing.
ScaleBITS enables automated, fine-grained bitwidth choice under a memory budget while preserving hardware speed. A new sensitivity study guides a hardware-aligned, block-wise weight split powered by bi-directional channel reordering. Global bitwidth choice becomes a constrained tuning problem, solved with a scalable take on the greedy method. Tests show large gains over uniform bit width and clear wins over sensitivity-aware baselines in the ultra-low-bit regime, with no added run-time overhead.
Catalog record
| CREATOR | Xinlin Li, Timothy Chou, Josh Fromm, Zichang Liu, Yunjie Pan, Christina Fragouli |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17698