arXiv · 2503.03874
LEWIS (LayEr WIse Sparsity) -- A Training Free Guided Model Merging Approach
Abstract
As specialized large language models (LLMs) become increasingly prevalent, model merging methods are being used to combine them to create a single multi-task model without requiring any additional data or training. However, these approaches fall short when the objective of merging is to increase the downstream model's performance on a particular task-specific benchmark. In this work, we propose LEWIS (Layer Wise Sparsity), a guided model-merging framework that uses activation-based layer importance to dynamically adjust layer-wise task-vector sparsity required for the merge process. LEWIS uses a calibration dataset to prioritize critical layers during the task-vector pruning process required for model merging. This approach guides existing merging methods by preserving essential layer-wise task-specific knowledge while ensuring the merged model performs the best at benchmarks resembling the calibration dataset. Our experiments demonstrate the effectiveness of LEWIS with performance improvements of code instruction-following and math-solving models created through model merging up to 4 percent and 11.3 percent, respectively, outperforming unguided data-less model merging approaches that use uniform-sparsity.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Hetarth Chopra, Vidhi Rambhia, Vikram Adve. 2025-03-05. LEWIS (LayEr WIse Sparsity) -- A Training Free Guided Model Merging Approach. https://arxiv.org/abs/2503.03874
Cite the original work for its findings. Save a collection to share your selection of sources.