I-SHEEP: Self-Alignment of LLM from Scratch through an Iterative Self-Enhancement Paradigm

TL;DR

I-SHEEP introduces an iterative self-enhancement paradigm enabling LLMs to self-align continuously from scratch, significantly improving their performance across multiple benchmarks compared to one-time alignment methods.

Contribution

This paper presents I-SHEEP, a novel human-like iterative self-alignment approach for LLMs that surpasses prior one-time alignment techniques in various tasks.

Findings

Achieves up to 78.2% improvement in Alpaca Eval

Surpasses base models in code generation, TrivialQA, and SQuAD tasks

Demonstrates continuous capacity enhancement over iterations

Abstract

Large Language Models (LLMs) have achieved significant advancements, however, the common learning paradigm treats LLMs as passive information repositories, neglecting their potential for active learning and alignment. Some approaches train LLMs using their own generated synthetic data, exploring the possibility of active alignment. However, there is still a huge gap between these one-time alignment methods and the continuous automatic alignment of humans. In this paper, we introduce \textbf{I-SHEEP}, an \textbf{I}terative \textbf{S}elf-En\textbf{H}anc\textbf{E}m\textbf{E}nt \textbf{P}aradigm.This human-like paradigm enables LLMs to \textbf{continuously self-align from scratch with nothing}. Compared to the one-time alignment method Dromedary \cite{sun2023principledriven}, which refers to the first iteration in this paper, I-SHEEP can significantly enhance capacities on both Qwen and Llama models. I-SHEEP achieves a maximum relative improvement of 78.2\% in the Alpaca Eval, 24.0\% in the MT Bench, and an absolute increase of 8.88\% in the IFEval accuracy over subsequent iterations in Qwen-1.5 72B model. Additionally, I-SHEEP surpasses the base model in various standard benchmark generation tasks, achieving an average improvement of 24.77\% in code generation tasks, 12.04\% in TrivialQA, and 20.29\% in SQuAD. We also provide new insights based on the experiment results. Our codes, datasets, and models are available at \textbf{https://anonymous.4open.science/r/I-SHEEP}.