Not Everything is All You Need: Toward Low-Redundant Optimization for Large Language Model Alignment

TL;DR

This paper introduces ALLO, a low-redundant alignment method for large language models that selectively optimizes key neurons related to human preferences, improving alignment efficiency and performance.

Contribution

The paper proposes a novel neuron selection strategy and a two-stage learning process to reduce redundancy and enhance LLM alignment with human preferences.

Findings

Selective neuron optimization improves alignment performance.

Reducing redundant neurons accelerates convergence.

ALLO outperforms baseline methods on multiple datasets.

Abstract

Large language models (LLMs) are still struggling in aligning with human preference in complex tasks and scenarios. They are prone to overfit into the unexpected patterns or superficial styles in the training data. We conduct an empirical study that only selects the top-10\% most updated parameters in LLMs for alignment training, and see improvements in the convergence process and final performance. It indicates the existence of redundant neurons in LLMs for alignment training. To reduce its influence, we propose a low-redundant alignment method named \textbf{ALLO}, focusing on optimizing the most related neurons with the most useful supervised signals. Concretely, we first identify the neurons that are related to the human preference data by a gradient-based strategy, then identify the alignment-related key tokens by reward models for computing loss. Besides, we also decompose the alignment process into the forgetting and learning stages, where we first forget the tokens with unaligned knowledge and then learn aligned knowledge, by updating different ratios of neurons, respectively. Experimental results on 10 datasets have shown the effectiveness of ALLO. Our code and data are available at \url{https://github.com/RUCAIBox/ALLO}.