WaterSIC: information-theoretically (near) optimal linear layer quantization

Egor Lifar, Semyon Savkin, Or Ordentlich, Yury Polyanskiy · 2026

Read WaterSIC: information-theoretically (near) optimal linear layer quantization by Egor Lifar, Semyon Savkin, Or Ordentlich, Yury Polyanskiy (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-2603-04956.txt.

Description

Egor Lifar and the rest of the team tackle converting a dense linear layer to low precision, analyzing compressed length against output discrepancy information-theoretically. They show the popular GPTQ algorithm can sit arbitrarily far from the information-theoretic limit. Their WaterSIC algorithm closes that gap to within 0.255 bits uniformly over all input activation covariance matrices, by giving other quantization rates to other weight-matrix columns, echoing classical waterfilling. Applied to Llama and Qwen models, it sets new state-of-the-art scores at every quantization rate from 1 to 4 bits.

Catalog record

CREATOREgor Lifar, Semyon Savkin, Or Ordentlich, Yury Polyanskiy
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE88497
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.04956

Browse the full library · About the Tower of Babel