Skip to content
Live
QbitAI·· 22 hours ago38

Shishi Technology releases multimodal model service, AI video reasoning speed increases 2.48 times without quality drop

Brief overview

Shishi Technology releases multimodal model service, through adaptive acceleration technology, cache management, modal token compression and Context Engine, the video generation reasoning speed is increased by 2.48 times while maintaining generation quality. The service combines SGLang to optimize GPU memory utilization, modal adaptive token compression accelerates 1.33 times in WorldSense evaluation, supports structured information connection of multimodal input, aiming to improve the efficiency of AI video large-scale production.

Source: QbitAI · qbitai.com