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

Large language models(LLMs) have sparked a new wave of exciting AIapplications. Hosting these models at scale requires significant memoryresources. One crucial memory bottleneck for the deployment stems from thecontext window. It is commonly recognized that model weights are memory hungry;however, the size of key-value embedding stored during the generation process(KV cache) can easily surpass the model size. The enormous size of the KV cacheputs constraints on the inference batch size, which is crucial for highthroughput inference workload. Inspired by an interesting observation of theattention scores, we hypothesize the persistence of importance: only pivotaltokens, which had a substantial influence at one step, will significantlyinfluence future generations. Based on our empirical verification andtheoretical analysis around this hypothesis, we propose Scissorhands, a systemthat maintains the memory usage of the KV cache at a fixed budget withoutfinetuning the model. In essence, Scissorhands manages the KV cache by storingthe pivotal tokens with a higher probability. We validate that Scissorhandsreduces the inference memory usage of the KV cache by up to 5X withoutcompromising model quality. We further demonstrate that Scissorhands can becombined with 4-bit quantization, traditionally used to compress model weights,to achieve up to 20X compression.

Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time