Upstage-NIA 'Open Ko-LLM Leaderboard' exceeds 100 models in 2 weeks
2023/10/12
UPSTAGE-NIA ESTABLISHED KOREAN LLM LEADERBOARD EXCEEDS 100 REGISTERED MODELS WITHIN 2 WEEKS OF LAUNCH
VARIOUS KOREAN LANGUAGE SPECIALIZED MODELS ARE COMPETING FOR SUCCESS FROM INDIVIDUALS TO COMPANIES... EXPECTED TO PLAY A CENTRAL ROLE IN KOREAN OPEN SOURCE LLM
LAYING THE FOUNDATION FOR STRENGTHENING KOREA’S AI COMPETITIVENESS BY EXPANDING THE DOMESTIC LLM ECOSYSTEM
(Upstage = 2023/10/12) The number of models registered on the Korean Large Language Model (LLM) evaluation leaderboard built by Upstage and the National Information Society Agency (NIA) exceeded 100 in two weeks.
Upstage (CEO Kim Seong-hoon) announced that the Korean LLM evaluation leaderboard 'Open Ko-LLM Leaderboard', which was launched on the 27th, achieved the success of exceeding 100 registered models in just two weeks after its release.
The Open Ko-LLM Leaderboard, built by Upstage in collaboration with NIA, is an open platform where anyone can register their own Korean LLM and compete with other models. In particular, it has the strength of being a Korean-specific leaderboard by building high-quality data that reflects the characteristics and culture of the Korean language.
The Open Ko-LLM leaderboard showed rapid expansion with over 100 models registered within two weeks of launch, despite the Chuseok holiday. Various companies and academic organizations, including individual researchers, SKT, Maum AI, 42Maru, Electronics and Telecommunications Research Institute (ETRI), and Korea University, are competing and attracting attention as the rankings fluctuate on a daily basis.
In particular, the Open Ko-LLM leaderboard is a collection of existing famous Korean open source models such as 'Ko-Alpaca', Korea University's 'KULLM (Cloud)', and 'Polyglot-Ko', showing the performance of Korean-specific LLM within the industry. It is an assessment that it has established itself as a barometer for evaluation. In particular, it received global attention, with Hugging Face's head of the Asia-Pacific region directly mentioning the Upstage leaderboard on social media.
In the future, Upstage plans to carry out various activities to revitalize the domestic LLM ecosystem, such as highlighting the top models by calculating monthly leaderboard rankings and holding open seminars by inviting researchers who developed the models. Researchers interested in participating in the leaderboard can check detailed information and participate in Hugging Face’s Open Ko-LLM Leaderboard space.
Upstage plans to present a blueprint for strengthening Korea's AI competitiveness through the operation of the Open Ko-LLM leaderboard. In particular, by operating a leaderboard based on Korean data in addition to the existing English-centered evaluation system, we are laying the foundation for expanding the Korean LLM ecosystem. In addition, it is expected to serve as a focal point for the domestic open source LLM camp by allowing various researchers to freely share each other's models and further advance technology through joint research and exchange.
Kim Seong-hoon, CEO of Upstage, said, “We are happy that the Open Ko-LLM leaderboard has reached a meaningful milestone of exceeding 100 models in two weeks,” and added, “In the future, Upstage will set a benchmark for Korean-specialized LLM performance and expand the domestic AI ecosystem.” “We will strive to strengthen our growth and global competitiveness,” he said.
Meanwhile, Upstage's LLM 'SOLAR' ranked first in the world as a result of exceeding ChatGPT's benchmark score on the open LLM leaderboard operated by Hugging Face last August. Recently, only four companies' LLMs, including OpenAI ChatGPT, Google Farm, Metarama, and Entropic Claude, were previously listed, and Solar was registered as the main model of Poe, which has become the standard for high-performance models, surprising the global market once again. did.