English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Despite their remarkable performance on abstractive summarization, largelanguage models (LLMs) face two significant challenges: their considerable sizeand tendency to hallucinate. Hallucinations are concerning because they erodethe reliability of LLMs and raise safety issues. Pruning is a technique thatreduces model size by removing redundant weights to create sparse models thatenable more efficient inference. Pruned models yield comparable performance totheir counterpart full-sized models, making them ideal alternatives whenoperating on a limited budget. However, the effect that pruning has uponhallucinations in abstractive summarization with LLMs has yet to be explored.In this paper, we provide an extensive empirical study on the hallucinationsproduced by pruned models across three standard summarization tasks, twopruning approaches, three instruction-tuned LLMs, and three hallucinationevaluation metrics. Surprisingly, we find that pruned LLMs hallucinate lesscompared to their full-sized counterparts. Our follow-up analysis suggests thatpruned models tend to depend more on the source input and less on theirparametric knowledge from pre-training for generation. This greater dependencyon the source input leads to a higher lexical overlap between generated contentand the source input, which can be a reason for the reduction inhallucinations.

Lighter, yet More Faithful: Investigating Hallucinations in Pruned Large Language Models for Abstractive Summarization