Sparsity-Aware Evolution for Model Merging
This work addresses model merging for large language models, offering an incremental improvement with a simple, orthogonal approach.
The paper tackles the problem of model merging by proposing a sparsity-aware evolutionary framework that uses iterative pruning-merging cycles as a mutation operator, and it demonstrates improved reliability across multiple large-scale LLM benchmarks.
We propose a sparsity-aware evolutionary (SAE) framework for model merging that involves iterative pruning-merging cycles to act as a novel mutation operator. We incorporate the sparsity constraints into the score function, which steers the evolutionary process to favor more sparse models, in addition to other conventional performance scores. Interestingly, the by-product of \textit{competition} for sparsity introduces an extra local \textit{attraction} and interplay into the evolutionary process: if one competitor has more zero elements, the other competitor's non-zero elements will occupy those positions, even though the less sparse competitor loses to the more sparse competitor in other positions. The proposed pipeline is evaluated on a variety of large-scale LLM benchmarks. Experiments demonstrate that our approach can improve model merging reliability across multiple benchmarks, and is easy to incorporate due to its simplicity and being orthogonal to most existing approaches.