CRaSh: Clustering, Removing, and Sharing Enhance Fine-tuning without Full Large Language Model

TL;DR

This paper introduces CRaSh, a training-free method that enhances offsite-tuning of large language models by exploiting modular structures, leading to improved emulator performance without full model fine-tuning.

Contribution

It reveals a modular layer structure in LLMs and proposes CRaSh, a novel strategy for better offsite-tuning performance without requiring full model training.

Findings

Modular structure emerges as model size increases.

CRaSh significantly improves offsite-tuning performance.

Optima from fine-tuning with and without full models are linearly connected.

Abstract

Instruction tuning has recently been recognized as an effective way of aligning Large Language Models (LLMs) to enhance their generalization ability across various tasks. However, when tuning publicly accessible, centralized LLMs with private instruction data, privacy concerns are inevitable. While direct transfer of parameterized modules between models is a plausible approach to address this, its implications and effectiveness need further exploration. This paper focuses on Offsite-Tuning (OFT), a representative technique that transfers transformer blocks between centralized LLMs and downstream emulators. Given the limited understanding of the underlying mechanism of OFT, we perform an empirical analysis on LLMs from the perspectives of representation and functional similarity. Interestingly, our findings reveal a unique modular structure within the layers of LLMs that appears to emerge as the model size expands. Simultaneously, we note subtle but potentially significant changes in representation and intermediate predictions across the layers. Inspired by these observations, we propose CRaSh, involving Clustering, Removing, and Sharing, a training-free strategy to derive improved emulators from LLMs. CRaSh significantly boosts performance of OFT with billions of parameters. Furthermore, we investigate the optimal solutions yielded by fine-tuning with and without full model through the lens of loss landscape. Our findings demonstrate a linear connectivity among these optima falling over the same basin, thereby highlighting the effectiveness of CRaSh and OFT. The source code is publicly available at https://github.com/TsinghuaC3I/CRaSh.