# Ultra-low latency vision-language

FlashVL is a new approach to optimizing vision-language models (VLMs) for real-time applications, aiming to achieve ultra-low latency and high throughput without sacrificing accuracy.