MERGE$^3$: Efficient Evolutionary Merging on Consumer-grade GPUs
This work addresses the problem of efficient evolutionary model merging for researchers and developers working with limited computational resources, particularly those in the natural language processing community.
The authors tackled the problem of evolutionary model merging on consumer-grade GPUs, achieving a 50x reduction in fitness computation costs while preserving performance. This resulted in state-of-the-art multilingual and cross-lingual merging with significantly lower computational overhead.
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.