Activation-Informed Merging of Large Language Models
This work addresses the challenge of efficiently combining fine-tuned LLMs for better task performance, though it is incremental as it builds on existing merging methods.
The paper tackled the problem of improving model merging for large language models by introducing Activation-Informed Merging (AIM), which integrates activation space information to enhance performance and robustness, resulting in up to a 40% increase in benchmark performance.
Model merging, a method that combines the parameters and embeddings of multiple fine-tuned large language models (LLMs), offers a promising approach to enhance model performance across various tasks while maintaining computational efficiency. This paper introduces Activation-Informed Merging (AIM), a technique that integrates the information from the activation space of LLMs into the merging process to improve performance and robustness. AIM is designed as a flexible, complementary solution that is applicable to any existing merging method. It aims to preserve critical weights from the base model, drawing on principles from continual learning (CL) and model compression. Utilizing a task-agnostic calibration set, AIM selectively prioritizes essential weights during merging. We empirically demonstrate that AIM significantly enhances the performance of merged models across multiple benchmarks. Our findings suggest that considering the activation-space information can provide substantial advancements in the model merging strategies for LLMs, with up to a 40% increase in benchmark performance.