An Empirical Study of SFT-DPO Interaction and Parameterization in Small Language Models

TL;DR

This study empirically compares supervised fine-tuning, DPO, and parameterization methods on small language models, revealing that full-parameter training remains most effective, with limited benefits from preference optimization and low-rank adaptation.

Contribution

It provides a systematic comparison of SFT, DPO, and parameterization techniques on small models, highlighting the dominance of full-parameter fine-tuning in this regime.

Findings

DPO offers small, task-dependent improvements over SFT.

Full fine-tuning outperforms LoRA at equivalent training depth.

Preference optimization and low-rank adaptation yield limited gains.

Abstract

Direct Preference Optimization (DPO) is widely used after supervised fine-tuning (SFT) to align language models, yet empirical behavior under small backbones and modest data is under-specified. We systematically compare SFT-only, DPO-only, and staged SFT-to-DPO training alongside full fine-tuning (FFT) versus LoRA on a GPT-2-scale decoder, evaluating paraphrase detection and Shakespearean sonnet continuation. DPO yields small, task-dependent gains over strong SFT and can match competitive SFT accuracy without a warm start when the preference construction closely parallels the supervised objective. In contrast, parameterization dominates: FFT consistently outperforms LoRA at matched training depth, and LoRA does not reduce wall-clock time on our hardware. These findings indicate that, in this small-scale regime, supervised full-parameter adaptation remains the primary performance lever, while preference optimization and low-rank adaptation provide limited marginal returns.