Beyond Under-Alignment: Atomic Preference Enhanced Factuality Tuning for Large Language Models
This work addresses hallucination in LLMs, a critical issue for reliable AI applications, by introducing a novel tuning method that improves factuality across domains, though it is incremental as it builds on existing preference learning approaches.
The paper tackles the problem of large language models generating factually erroneous responses (hallucinations) by evaluating and improving factuality under distribution shifts, proposing APEFT, which enhances factuality awareness at the individual fact level and improves model performance by an average of 3.45% on both in-domain and out-of-domain datasets.
Large language models (LLMs) have achieved remarkable success but still tend to generate factually erroneous responses, a phenomenon known as hallucination. A recent trend is to use preference learning to fine-tune models to align with factuality. However, existing work primarily evaluates fine-tuned models on in-domain (ID) datasets and the factuality on out-of-domain (OOD) datasets remains underexplored. In this paper, we conduct a comprehensive evaluation of the factuality of different models tuned by various preference learning algorithms and demonstrate that their performance on OOD datasets either increases minimally or decreases. Subsequently, we reveal that the main cause of model's failure to uphold factuality under a distribution shift is \textbf{under-alignment}, rather than \textbf{over-alignment}, by analyzing the token distribution shift of the models before and after tuning. Finally, we propose \textbf{APEFT} (\textbf{A}tomic \textbf{P}reference \textbf{E}nhanced \textbf{F}actuality \textbf{T}uning), a framework that enhances model's awareness of factuality at the granularity of individual facts. Extensive experiments demonstrate that APEFT improves model performance by an average of $\boldsymbol{3.45\%}$ on both ID and OOD datasets, which is highly effective.