An Empirical Study of SFT-DPO Interaction and Parameterization in Small Language Models
This is an incremental study clarifying practical trade-offs for researchers working with small-scale language model alignment.
The study investigated how Direct Preference Optimization (DPO) interacts with supervised fine-tuning (SFT) in small language models, finding that DPO provides only small, task-dependent gains over SFT and that full fine-tuning consistently outperforms LoRA parameterization.
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.