English

MERGE$^3$: Efficient Evolutionary Merging on Consumer-grade GPUs

Neural and Evolutionary Computing 2025-05-12 v4 Artificial Intelligence Machine Learning

Abstract

Evolutionary model merging enables the creation of high-performing multi-task models but remains computationally prohibitive for consumer hardware. We introduce MERGE3^3, an efficient framework that makes evolutionary merging feasible on a single GPU by reducing fitness computation costs 50×\times while preserving performance. MERGE3^3 achieves this by Extracting a reduced dataset for evaluation, Estimating model abilities using Item Response Theory (IRT), and Evolving optimal merges via IRT-based performance estimators. Our method enables state-of-the-art multilingual and cross-lingual merging, transferring knowledge across languages with significantly lower computational overhead. We provide theoretical guarantees and an open-source library, democratizing high-quality model merging.

Keywords

Cite

@article{arxiv.2502.10436,
  title  = {MERGE$^3$: Efficient Evolutionary Merging on Consumer-grade GPUs},
  author = {Tommaso Mencattini and Adrian Robert Minut and Donato Crisostomi and Andrea Santilli and Emanuele Rodolà},
  journal= {arXiv preprint arXiv:2502.10436},
  year   = {2025}
}

Comments

In Proceedings of The Forty-Second International Conference on Machine Learning (ICML 2025)

R2 v1 2026-06-28T21:44:52.162Z