Physical AI HBM Smart Factory SDV AIoT Power Semicon 특수 가스 정정·반론보도 모음 e4ds plus

Nota Takes First Place Overall at NVIDIA's 'Nemotron Hackathon'... Demonstrates Potential for Data-Centric AI Optimization

Google 우선 소스 기사입력2026.04.27 09:36



Ranked 1st overall with synthetic data-based MoE quantization

Nota took first place overall at the 'Nemotron Hackathon' hosted by NVIDIA with its synthetic data-based MoE quantization technology.

According to Nota on the 24th, the company won first place in Track C of the competition with synthetic data generation technology for quantization specialized in the Mixture of Experts (MoE) structure, and took the overall championship by receiving the highest evaluation among all 20 participating teams.

The Nemotron Hackathon is an event designed to test practical problem-solving skills using NVIDIA's open-source AI model, 'Nemotron,' and consists of three tracks: AI agent development, model refinement, and data pipeline design. Participants performed tasks simulating actual industrial applications and were evaluated on the completeness of their technical skills.

Nota built a dataset suitable for MoE model quantization using a Nemotron 3 Super 120B-based agent. Notably, it approached the task by meticulously designing the data structure and quality to minimize model performance degradation. This is considered a case that utilizes the data itself as an optimization element, taking a different approach from conventional formula or algorithm-centric optimization.

This approach can also be interpreted as a trend demonstrating that the focus of AI optimization is expanding from model compression techniques to data design. Analysis suggests that these results confirm that data quality can actually play a crucial role in maintaining performance during the model lightweighting process.

Nota is currently applying NVIDIA technology to its Vision Language Model-based image analysis solution to implement capabilities for detecting and summarizing anomalies, utilizing this to improve response speeds in industrial settings. Building on this achievement, the company plans to expand the scope of application for related technologies.

Chae Myeong-su, CEO of Nota, stated, “This is a case that confirmed the importance of data design in the AI optimization process,” adding, “We will continue to develop technologies applicable in real-world environments.”