Towards a Holistic Evaluation of LLMs on Factual Knowledge Recall

TL;DR

This paper introduces FACT-BENCH, a comprehensive benchmark for evaluating LLMs' ability to recall factual knowledge, revealing impacts of instruction tuning, model size, and exemplars on knowledge recall performance.

Contribution

The paper presents FACT-BENCH, a new benchmark for holistic evaluation of LLMs' factual recall, and provides insights into factors affecting knowledge retention and the effects of fine-tuning strategies.

Findings

Instruction-tuning reduces factual recall performance.

Larger models outperform smaller ones across families.

Fine-tuning on known knowledge improves recall.

Abstract

Large language models (LLMs) have shown remarkable performance on a variety of NLP tasks, and are being rapidly adopted in a wide range of use cases. It is therefore of vital importance to holistically evaluate the factuality of their generated outputs, as hallucinations remain a challenging issue. In this work, we focus on assessing LLMs' ability to recall factual knowledge learned from pretraining, and the factors that affect this ability. To that end, we construct FACT-BENCH, a representative benchmark covering 20 domains, 134 property types, 3 answer types, and different knowledge popularity levels. We benchmark 31 models from 10 model families and provide a holistic assessment of their strengths and weaknesses. We observe that instruction-tuning hurts knowledge recall, as pretraining-only models consistently outperform their instruction-tuned counterparts, and positive effects of model scaling, as larger models outperform smaller ones for all model families. However, the best performance from GPT-4 still represents a large gap with the upper-bound. We additionally study the role of in-context exemplars using counterfactual demonstrations, which lead to significant degradation of factual knowledge recall for large models. By further decoupling model known and unknown knowledge, we find the degradation is attributed to exemplars that contradict a model's known knowledge, as well as the number of such exemplars. Lastly, we fine-tune LLaMA-7B in different settings of known and unknown knowledge. In particular, fine-tuning on a model's known knowledge is beneficial, and consistently outperforms fine-tuning on unknown and mixed knowledge. We will make our benchmark publicly available.