Insights into Alignment: Evaluating DPO and its Variants Across Multiple Tasks
This work provides insights into the effectiveness and limitations of alignment methods for LLMs, which is important for researchers and practitioners in AI, though it is incremental as it builds on existing DPO approaches.
This study evaluated Direct Preference Optimization (DPO) and its variants for aligning Large Language Models with human preferences across 13 benchmarks, finding that alignment methods achieve near-optimal performance with smaller training data, enhance mathematical problem-solving, and improve truthfulness when using instruction-tuned models.
This study evaluates Direct Preference Optimization (DPO) and its variants for aligning Large Language Models (LLMs) with human preferences, testing three configurations: (1) with Supervised Fine Tuning (SFT), (2) without SFT, and (3) without SFT but using an instruction tuned model. We further investigate how training set size influences model performance. Our evaluation spans 13 benchmarks covering dialogue, reasoning, mathematical problem-solving, question answering, truthfulness, MT-Bench, Big Bench, and the Open LLM Leaderboard. We find that: (1) alignment methods often achieve near optimal performance even with smaller subsets of training data; (2) although they offer limited improvements on complex reasoning tasks, they enhance mathematical problem-solving; and (3) using an instruction tuned model improves truthfulness. These insights highlight the conditions under which alignment methods excel, as well as their limitations.