DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models
This addresses the issue of unreliable fact generation in LLMs for users needing accurate information, though it is an incremental improvement as it builds on existing layer-localization insights.
The paper tackled the problem of hallucinations in large language models (LLMs) by proposing a decoding strategy that contrasts layer outputs to surface factual knowledge, resulting in improved truthfulness, such as a 12-17% absolute point increase on TruthfulQA for LLaMA models.
Despite their impressive capabilities, large language models (LLMs) are prone to hallucinations, i.e., generating content that deviates from facts seen during pretraining. We propose a simple decoding strategy for reducing hallucinations with pretrained LLMs that does not require conditioning on retrieved external knowledge nor additional fine-tuning. Our approach obtains the next-token distribution by contrasting the differences in logits obtained from projecting the later layers versus earlier layers to the vocabulary space, exploiting the fact that factual knowledge in an LLMs has generally been shown to be localized to particular transformer layers. We find that this Decoding by Contrasting Layers (DoLa) approach is able to better surface factual knowledge and reduce the generation of incorrect facts. DoLa consistently improves the truthfulness across multiple choices tasks and open-ended generation tasks, for example improving the performance of LLaMA family models on TruthfulQA by 12-17% absolute points, demonstrating its potential in making LLMs reliably generate truthful facts.