Mergenetic: a Simple Evolutionary Model Merging Library

2025-05-16

AI Summary

Existing research on language model merging lacks an efficient framework supporting flexible integration of merging strategies and evolutionary algorithms. This paper introduces the first open-source evolutionary model merging framework tailored for large language models (LLMs), enabling training-free, low-overhead co-optimization of merging policies and evolutionary operators on a single consumer-grade GPU (e.g., RTX 4090). Our key contributions are: (1) a lightweight low-rank fitness proxy evaluator that drastically reduces evaluation costs across multi-task and multilingual benchmarks; (2) a modular architecture compatible with mainstream weighted merging methods (e.g., TIES, DARE) and evolutionary strategies (e.g., genetic algorithms); and (3) state-of-the-art merging performance, thereby addressing the critical gap in scalable, easily deployable evolutionary LLM merging tools.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Model merging allows combining the capabilities of existing models into a new one - post hoc, without additional training. This has made it increasingly popular thanks to its low cost and the availability of libraries that support merging on consumer GPUs. Recent work shows that pairing merging with evolutionary algorithms can boost performance, but no framework currently supports flexible experimentation with such strategies in language models. We introduce Mergenetic, an open-source library for evolutionary model merging. Mergenetic enables easy composition of merging methods and evolutionary algorithms while incorporating lightweight fitness estimators to reduce evaluation costs. We describe its design and demonstrate that Mergenetic produces competitive results across tasks and languages using modest hardware.

Problem

• Lacks framework for evolutionary model merging in language models
• Needs efficient fitness estimation to reduce evaluation costs
• Requires flexible merging methods for diverse tasks and languages

Innovation

• Open-source library for evolutionary model merging
• Combines merging methods with evolutionary algorithms
• Uses lightweight fitness estimators to cut costs