Calibrating Beyond English: Language Diversity for Better Quantized Multilingual LLMs

Everlyn Asiko Chimoto, Mostafa Elhoushi, Bruce Bassett · 2026

Read Calibrating Beyond English: Language Diversity for Better Quantized Multilingual LLMs by Everlyn Asiko Chimoto, Mostafa Elhoushi, Bruce Bassett (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-2601-18306.txt.

Description

Everlyn Asiko Chimoto, Mostafa Elhoushi, and Bruce Bassett argue that calibration data for shrinking models should span many languages. Their 2026 preprint deals with quantization, which cuts a model's storage and compute costs. Quantization often costs accuracy. The standard post-training methods calibrate on small English-only sets. What that choice does to multilingual models has been little studied. The authors test language diversity in calibration for quantized multilingual models.

If diverse calibration guards non-English skill, the fix is cheap and broad. Most users of these models do not work in English alone. The study gives deployment teams a concrete lever to test. It also warns against English-only habits in model tooling.

Catalog record

CREATOREverlyn Asiko Chimoto, Mostafa Elhoushi, Bruce Bassett
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-18306

Browse the full library · About the Tower of Babel