| Mobile Web

Dinotisia unveils KV cache compression technology to ease AI computing bottlenecks

AI infrastructure company Dinotisia said on Wednesday it has released a paper and source code for KV cache compression technology that stores previously read context on GPUs to avoid recalculation by large language models. The company said its STAR-KV technology, a joint effort involving UC San Diego’s VVIP Lab and Dinotisia researchers, was selected as an ICML 2026 Spotlight paper. It said tests showed up to 75 percent reduction and up to 20-fold compression.