# English Calibration Erodes Lithuanian in Expert-Pruned Mixture-of-Experts Language Models

DevFeed: [English Calibration Erodes Lithuanian in Expert-Pruned Mixture-of-Experts Language Models](<https://devfeed.tech/articles/english-calibration-erodes-lithuanian-in-expert-pruned-mixture-of-experts-language-models-120642.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07654>)

Author: Tomas Rasymas, Povilas Kvedaras

Published: 2026-11-06T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [Model Optimizer](<https://devfeed.tech/topics/model-optimizer.md>)

Tags: [1-7](<https://devfeed.tech/tags/1-7.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [english](<https://devfeed.tech/tags/english.md>), [experts](<https://devfeed.tech/tags/experts.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [languages](<https://devfeed.tech/tags/languages.md>), [layers](<https://devfeed.tech/tags/layers.md>), [mixture-of-experts](<https://devfeed.tech/tags/mixture-of-experts.md>), [mixture-of-experts-moe](<https://devfeed.tech/tags/mixture-of-experts-moe.md>), [modeling](<https://devfeed.tech/tags/modeling.md>), [modelling](<https://devfeed.tech/tags/modelling.md>), [models](<https://devfeed.tech/tags/models.md>), [moe](<https://devfeed.tech/tags/moe.md>), [points](<https://devfeed.tech/tags/points.md>), [pruning](<https://devfeed.tech/tags/pruning.md>), [reading](<https://devfeed.tech/tags/reading.md>), [routing](<https://devfeed.tech/tags/routing.md>), [translation](<https://devfeed.tech/tags/translation.md>)

## AI overview

Across four MoE models pruned with REAP, English calibration transferred poorly to Lithuanian: at 25% pruning, median comprehension loss was 21.4 points for Lithuanian versus 1.7 for English. The transfer was asymmetric: Lithuanian calibration removed most of the Lithuanian loss and did not significantly hurt English comprehension, which averaged 2.5 points better than with English calibration; English language modeling was slightly worse. Routing offers a possible explanation: in the last layers, Lithuanian concentrated 55-72% of expert selections on eight experts, compared with 19-39% for English, and English calibration removed some of them. The abstract recommends calibrating and evaluating compressed models in every language they must serve.

## Source excerpt

arXiv:2611.07654v1 Announce Type: cross Abstract: Expert pruning compresses Mixture-of-Experts (MoE) language models by removing the experts that a calibration set marks as unimportant. Studies of high-resource languages largely find that English calibration transfers well to other languages. We test this for Lithuanian: we pruned four MoE models with REAP, removing 25% or 50% of the experts with English, Lithuanian or mixed calibration, and evaluated both languages on reading comprehension, translation and language modelling. English calibration does not transfer to Lithuanian: at 25% pruning, English comprehension fell by a median of 1.7 points but Lithuanian by 21.4. The transfer is asymmetric. Lithuanian calibration removed most of the Lithuanian loss and never significantly hurt English comprehension, which was on average 2.5 points better than under English calibration; only English language modeling got slightly worse. Routing suggests why: in the last layers, Lithuanian sends 55-72% of its expert selections to just eight experts, against 19-39% for English, and English calibration removes some of these experts. Compressed models should be calibrated, and evaluated, in every language they must serve.