Hark describes scaling multimodal data curation with SGLang and KubeRay

Hark engineers Kai-Hsun Chen and Cindy Zhang describe how the company scales multimodal data curation using SGLang and KubeRay, according to a post from the LMSYS Org account on X. The post covers GPU orchestration and faster startup for inference workers.

The reports do not include figures for throughput, cost, or startup time, so the post's performance claims cannot be checked from the available material.