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

To reduce the computation cost and the energy consumption in large languagemodels (LLM), skimming-based acceleration dynamically drops unimportant tokensof the input sequence progressively along layers of the LLM while preservingthe tokens of semantic importance. However, our work for the first time revealsthe acceleration may be vulnerable to Denial-of-Service (DoS) attacks. In thispaper, we propose No-Skim, a general framework to help the owners ofskimming-based LLM to understand and measure the robustness of theiracceleration scheme. Specifically, our framework searches minimal andunnoticeable perturbations at character-level and token-level to generateadversarial inputs that sufficiently increase the remaining token ratio, thusincreasing the computation cost and energy consumption. We systematicallyevaluate the vulnerability of the skimming acceleration in various LLMarchitectures including BERT and RoBERTa on the GLUE benchmark. In the worstcase, the perturbation found by No-Skim substantially increases the runningcost of LLM by over 145% on average. Moreover, No-Skim extends the evaluationframework to various scenarios, making the evaluation conductible withdifferent level of knowledge.

No-Skim: Towards Efficiency Robustness Evaluation on Skimming-based Language Models