Towards Robust Calibration Dataset for Domain-General Mixture of Experts Pruning

Abstract:Post-training expert pruning is a promising deployment technique for Mixture-of-Experts (MoE) models. The calibration set is an integral part of the pruning process, however, its correlation with evaluation performance is largely misunderstood and unintuitive. In this work, we present an empirical study of the robustness of calibration perturbations, specifically in MoE models. Showing how calibration size, language, and data distribution affect layer-wise expert pruning for gpt-oss-20b. Our findings reveal that (1) domain-general pruning quality receives minimal benefit from scaling calibration data beyond small sample sizes, (2) cross-lingual calibration transfers effectively, with English-only calibration maintaining strong performance across seven languages of varying typologies and scripts, and (3) a diverse calibration data resembling the pretraining distribution can yield comparable or superior performance to test-distribution data. Ultimately, we provide practical insights as a recipe for better understanding calibration for more effective domain-general expert pruning.

Computation and Language (cs.CL)

arXiv:2610.15578 [cs.CL]

(or arXiv:2610.15578v1 [cs.CL] for this version)

https://doi.org/10.48550/arXiv.2610.15578 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.