The Day After Tomorrow, Open Source Release of Self-Developed Korean LLM Foundation Model 'Motif'

Utilized for Specialized Document Learning
"Contributing to the Development of the Korean AI Ecosystem"

Artificial intelligence (AI) infrastructure solution company 'More' announced on the 3rd the open-source release of its self-developed Korean large language model (LLM) foundation model, Llama-3-Motif-102B (hereinafter Motif).

Making the Korean LLM foundation model and its source code available for anyone to use holds great significance for the development of the domestic AI industry. Securing AI capabilities specialized in the native language is an essential element of sovereign AI.

More plans to contribute to the growth of the Korean AI ecosystem by distributing a high-performance Korean LLM as open source on Hugging Face. To this end, it will also establish a subsidiary to fully promote its AI model business.

More stated, "Motif is a Korean LLM with 102 billion parameters," adding, "It scored 64.74 on the 'KMMLU' benchmark, a Korean AI performance evaluation system, demonstrating superior Korean language processing performance compared to LLMs from Meta, Google, and Naver."

Motif utilized not only web-collected texts but also publicly available specialized documents such as domestic patents and research reports as training data. Additionally, it secured the largest-scale refined Korean language dataset domestically and included it in the training. Two versions of open-source models are released: a pre-trained language model and an instruct model specialized in following instructions.

Jangwon Cho, CEO of More, said, "Releasing a high-performance LLM as open source for anyone to use is to contribute to the growth of the domestic AI ecosystem and sovereign AI," and added, "We hope many companies, including startups striving for AI industry development, actively utilize it."

Based on its AI infrastructure software technology, More plans to develop LLMs specialized in fields such as intellectual property (IP) industries, as well as medical, legal, and financial sectors, and to pursue the development of multimodal models, aiming to become an AI model hub.