Nota, an artificial intelligence (AI) model slimming and optimization corporations, said on the 30th that it successfully optimized LG AI Research Institute's national flagship AI model "K-EXAONE 236B" on FuriosaAI's data center Neural Processing Unit (NPU).
K-EXAONE 236B is a large AI model with about 236 billion parameters, and it adopts a mixture-of-experts (MoE) architecture that selectively leverages multiple expert models. While the MoE structure has the advantage of improving the efficiency of large models, the optimization process requires sophisticated techniques to ensure that each expert model operates stably.
In this project, Nota optimized K-EXAONE for FuriosaAI's data center NPU environment. The company said it reduced the model size by about 71%, lowering the memory burden required to run large AI models, while maintaining accuracy at a level similar to the original model on key evaluations such as scientific reasoning, instruction following, and math problem solving.
A Nota official said, "We minimized performance loss by precisely analyzing sections where degradation could occur and applying optimization only where needed," adding, "This means we optimized a 236 billion-parameter model to run more efficiently, showing the potential to improve the operational efficiency of data center AI infrastructure."
In the global AI industry recently, access to cutting-edge AI models and the infrastructure that powers them has emerged as a key issue. In particular, following export control discussions surrounding some AI models and infrastructure, including Anthropic's "Mythos," demand has grown for sovereign AI as countries move to secure domestic AI models and computing infrastructure.
Chae Myeong-su, CEO of Nota, said, "What matters amid the attention on sovereign AI is that models, semiconductors, and optimization software are linked into a single, workable AI infrastructure," adding, "This achievement is a case that confirms the real-world operability of large AI models through the combination of FuriosaAI's data center NPU, LG's national flagship AI model K-EXAONE, and Nota's optimization technology."