aiqbench analysis
The paper tests whether pruning redundant attention heads in GPT-2-XL reduces verbatim memorisation at inference time. The reported 91.5% drop in memorised-sequence reproduction is a single-model result on known memorised sequences, so generalisation to other models and to real privacy or IP leakage remains open. Readers should watch whether the perplexity gain holds beyond the benchmark used to select the heads.
Fewer Heads, Fewer Memories: Pruning Redundant Attention Heads Suppresses Verbatim Memorisation
arXiv:2610.40974v1 Announce Type: cross Abstract: Large language models have been shown to memorise and reproduce verbatim sequences from their training data, raising concerns about privacy leakage, intellectual property exposure, and the limits of generalisation. Recent theoretical work suggests that attention heads may contribute to this behaviour by functioning as associative memory mechanisms, with memorisation capacity scaling with the number of heads. At the same time, empirical work has shown that many attention heads are redundant and can be pruned with limited impact on downstream performance. Despite these parallel findings, the effect of pruning redundant heads on memorised-sequence reproduction remains unexplored. In this work, we demonstrate that zeroing out a substantial fraction of attention heads prior to inference can significantly suppress verbatim memorisation while preserving, or even improving, language modelling performance on the benchmark used to identify redundant heads. Evaluating GPT-2-XL(1.5B) on known memorised sequences, we find that pruning 729 task-redundant heads reduces the memorisation count by 91.5%, while improving WikiText-103 perplexity by 0.92. These findings suggest that attention-head redundancy can be exploited to mitigate memorisation at inference time, without retraining or modifying the training data.
Abstract:Large language models have been shown to memorise and reproduce verbatim sequences from their training data, raising concerns about privacy leakage, intellectual property exposure, and the limits of generalisation. Recent theoretical work suggests that attention heads may contribute to this behaviour by functioning as associative memory mechanisms, with memorisation capacity scaling with the number of heads. At the same time, empirical work has shown that many attention heads are redundant and can be pruned with limited impact on downstream performance. Despite these parallel findings, the effect of pruning redundant heads on memorised-sequence reproduction remains unexplored. In this work, we demonstrate that zeroing out a substantial fraction of attention heads prior to inference can significantly suppress verbatim memorisation while preserving, or even improving, language modelling performance on the benchmark used to identify redundant heads. Evaluating GPT-2-XL(1.5B) on known memorised sequences, we find that pruning 729 task-redundant heads reduces the memorisation count by 91.5%, while improving WikiText-103 perplexity by 0.92. These findings suggest that attention-head redundancy can be exploited to mitigate memorisation at inference time, without retraining or modifying the training data.
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2610.40974 [cs.CL]
(or arXiv:2610.40974v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.40974 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 278 words