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

TL;DR

MERGE$^3$ is a novel framework that enables efficient evolutionary model merging on consumer-grade GPUs by significantly reducing computational costs while maintaining high performance, facilitating broader access to multi-task model development.

Contribution

It introduces a GPU-efficient evolutionary merging method using IRT-based estimators, reducing costs by 50x and enabling high-quality multilingual model merging.

Findings

Reduces fitness computation costs 50 times

Enables effective multilingual and cross-lingual merging

Provides theoretical guarantees and open-source tools

Abstract

Evolutionary model merging enables the creation of high-performing multi-task models but remains computationally prohibitive for consumer hardware. We introduce MERGE$^3$, an efficient framework that makes evolutionary merging feasible on a single GPU by reducing fitness computation costs 50$\times$ while preserving performance. MERGE$^3$ 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.