DEFT: Distribution-guided Efficient Fine-Tuning for Human Alignment

TL;DR

DEFT is a novel fine-tuning framework that improves human alignment and generalization of language models by using distributional guidance and data filtering, reducing training time.

Contribution

The paper introduces DEFT, a new alignment method that enhances efficiency and performance by leveraging differential distribution rewards and data filtering.

Findings

DEFT outperforms baseline methods in alignment capability.

DEFT reduces training time significantly.

Models fine-tuned with DEFT maintain better generalization.

Abstract

Reinforcement Learning from Human Feedback (RLHF), using algorithms like Proximal Policy Optimization (PPO), aligns Large Language Models (LLMs) with human values but is costly and unstable. Alternatives have been proposed to replace PPO or integrate Supervised Fine-Tuning (SFT) and contrastive learning for direct fine-tuning and value alignment. However, these methods still require voluminous data to learn preferences and may weaken the generalization ability of LLMs. To further enhance alignment efficiency and performance while mitigating the loss of generalization ability, this paper introduces Distribution-guided Efficient Fine-Tuning (DEFT), an efficient alignment framework incorporating data filtering and distributional guidance by calculating the differential distribution reward based on the output distribution of language model and the discrepancy distribution of preference data. A small yet high-quality subset is filtered from the raw data using a differential distribution reward, which is then incorporated into existing alignment methods to guide the model's output distribution. Experimental results demonstrate that the methods enhanced by DEFT outperform the original methods in both alignment capability and generalization ability, with significantly reduced training time.