Nota, Upstage Consortium Advances to Third Round of Independent Foundation Model

This page was machine-translated and may differ from the original. View original

Nota, Upstage Consortium Advances to Third Round of Independent Foundation Model

Published2026.08.20 09:35

Solar Open 2 Optimization: H100 GPU 8 Units→2 Units Operation, 76.5% Memory Reduction

Model requiring 8 H100 GPUs now runs on 2 units; 80,000 cumulative downloads achieved on Hugging Face. Nota AI, an AI model optimization and lightweight company, has entered the third round of the government's independent AI Foundation Model (DokPaMo) initiative as a member of the Upstage Consortium, and is now actively providing infrastructure efficiency technology support for domestic AI models.

On the 20th, Nota announced that following the DokPaMo Phase 2 evaluation, the three-company consortium of Upstage, SK Telecom, and LG AI Research Center was selected to advance to the third round, and will continue to serve its role in optimization and lightweight operations within the consortium.

Nota is responsible for reducing computational resources and memory usage so that Upstage's AI Foundation Model 'Solar' series can operate efficiently across various domestic hardware and service environments.

Solar Open 2 Optimization Results

Nota released the optimized version of Upstage's 'Solar Open 2' on Hugging Face ahead of the Phase 2 evaluation results announcement.

Solar Open 2 is a model with 250 billion parameters. Nota applied the approach of converting weights to 4-bit format based on INT4 or NVFP4 and selectively removing computation modules with low importance.

As a result, memory required for weight storage was reduced from 500.6GB to 117.8GB, a 76.5% reduction, and the original model that previously required 8 NVIDIA H100 GPUs can now run on 2 units.

The model recorded 80,000 cumulative downloads on Hugging Face.

DokPaMo Initiative Background and Evaluation Criteria

DokPaMo is a government initiative to develop and foster foundation models with global competitiveness based on independent domestic technology.

In this Phase 2 evaluation, comprehensive assessment was conducted based on △quantitative model performance △resource utilization efficiency △usability and practicality △industry and service scalability.

The Upstage Consortium reportedly received positive evaluation from the expert committee for presenting the possibility of combining domestic AI software and hardware ecosystems through service implementation linked with domestic platforms and utilization of FuriosaAI NPU.

Chae Myung-soo, CEO of Nota, stated, "We will continuously advance optimization technologies so that national representative AI models can be utilized cost-effectively across diverse industrial sites," and added, "We will do our best in the DokPaMo third round evaluation together with Upstage."

#독파모 #AI 모델 경량화 #Nota AI #Solar Open 2 #H100 GPU #INT4 양자화 #메모리 최적화 #FuriosaAI NPU

To request a correction, reply or follow-up report on this article, see how to file a request. Previously published statements are collected in corrections & replies.