AI model lightweighting and optimisation company Nota said on Thursday it took the overall top prize among 20 teams at the Nvidia Nemotron hackathon with its synthetic data generation technology specialised for mixture-of-experts (MoE) quantisation.

The hackathon was held to share research results from Nvidia's open-source AI model Nemotron and to enhance the practical application capabilities of domestic developers. It ran in three tracks: AI agent development, advancement of domain-specific models and synthetic data pipeline design.

Nota used an agent based on Nemotron 3 Super 120B to build a dataset for quantisation tailored to the MoE architecture. While conventional quantisation has focused on algorithm-centred optimisation, Nota explained it improved performance through a data-centric approach that designs dataset structure, quality and fitness for purpose.

Myungsoo Chae (채명수), chief executive of Nota, said: "This award is an achievement that shows we can create new possibilities in AI optimisation depending on how we design and use data fit for the purpose." He added: "Based on cooperation with Nvidia, we will continue to advance data-centric AI optimisation technology."