Saved in:
Bibliographic Details
Main Authors: Nobari, Amin Heyrani, Alim, Kaveh, ArjomandBigdeli, Ali, Srivastava, Akash, Ahmed, Faez, Azizan, Navid
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.02421
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915603254607872
author Nobari, Amin Heyrani
Alim, Kaveh
ArjomandBigdeli, Ali
Srivastava, Akash
Ahmed, Faez
Azizan, Navid
author_facet Nobari, Amin Heyrani
Alim, Kaveh
ArjomandBigdeli, Ali
Srivastava, Akash
Ahmed, Faez
Azizan, Navid
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2502_02421
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Activation-Informed Merging of Large Language Models
Nobari, Amin Heyrani
Alim, Kaveh
ArjomandBigdeli, Ali
Srivastava, Akash
Ahmed, Faez
Azizan, Navid
Computation and Language
Artificial Intelligence
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.
title Activation-Informed Merging of Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2502.02421