Within the Sovereign AI Foundation Model Project, Nota optimized large-scale LLMs to run on fewer GPUs through model compression, while using Backend.AI and FastTrack to streamline experimentation and resource management. This enabled serving a 250B-scale model with significantly reduced GPU requirements while maintaining quality, and extending on-device techniques to data center scale.

ML research scientist Eunseong Choi explains how using Backend.AI on Sungkyunkwan University’s on‑premise supercomputing center reduced the overhead of managing GPU infrastructure and gave his lab a consistent multi‑GPU environment to focus on research.

The Upstage consortium, with Lablup as infrastructure partner, passed Phase 1 of Korea's Sovereign AI Foundation Model program. Team members share the behind-the-scenes story.