3BASiL: An Algorithmic Framework for Sparse plus Low-Rank Compression of LLMs

Mehdi Makni, Xiang Meng, Rahul Mazumder · 2026

Read 3BASiL: An Algorithmic Framework for Sparse plus Low-Rank Compression of LLMs by Mehdi Makni, Xiang Meng, Rahul Mazumder (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-2603-01376.txt.

Description

Makni, Meng, and Mazumder present a 2026 paper on shrinking large language models by splitting each weight matrix into a sparse part plus a low-rank part. Their method, 3BASiL-TM, tunes the split layer by layer, then refines all layers together so compressed outputs stay close to the original. Tests report the WikiText2 perplexity gap for a compressed LLaMA-8B model cut by over 30 percent.

Catalog record

CREATORMehdi Makni, Xiang Meng, Rahul Mazumder
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE114637
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.01376

Browse the full library · About the Tower of Babel