English < Article List - DigitalToday

Search results for PolarQuant

AI & Enterprise

KAIST professor involved in TurboQuant development says it will be core basis for running large-scale AI models

Google Research and others have unveiled TurboQuant, a technology aimed at easing AI bottlenecks by improving memory efficiency by up to six times. KAIST said Han Insoo of its Department of Electrical Engineering participated in the work with a joint team from Google Research, DeepMind and New York University. KAIST said TurboQuant compresses internal model information with little accuracy loss and addresses memory bottlenecks during AI inference.

AI & Enterprise

Google unveils TurboQuant AI compression algorithm cutting memory use sixfold

Google has unveiled TurboQuant, a new compression algorithm designed to cut memory use and increase speed for large language models. The technology targets the key-value cache that stores important information in LLMs, aiming to reduce memory use while maintaining performance and accuracy. Google said early tests showed memory use fell by up to six times and performance improved by up to eight times under some benchmarks.