The Large Language Model (LLM) developed by MOREH, an AI infrastructure solution company, ranked first globally with a high score of 77.29 points in the ‘Open LLM Leaderboard’ evaluation run by Hugging Face, the world’s largest machine learning platform.
'Open LLM Leaderboard' is a development platform used as a performance verification window for global AI models. Currently, over 3,400 models have been registered and are competing for ranking based on six evaluation indicators, including reasoning, common sense ability, and language understanding ability.
MOREH's latest model, 'MoMo-70B', which ranked first this time, is a huge language model with 70 billion parameters. This model first rose to the top of the leaderboard on January 17th and remains there to this day. This is a great achievement made in just three months from the start of model development to taking first place.
The reason MOREH was able to stand out as the world's No. 1 in the LLM field in such a short period of time was because it was supported by the technological prowess of ‘MoAI’, an AI platform developed and owned by MOREH. In order to quickly develop a top-level LLM, it is important to secure an efficient model learning method, and More's ‘MoAI’ platform helps to efficiently develop and learn large-scale AI models through highly parallel processing techniques.
Lim Junghwan, head of MOREH's AI Group said, "This is an achievement made by combining MOREH's outstanding AI model development ability and AI platform technology." He added, "Through 'MoAI', we were able to dramatically reduce the time and cost required to develop and learn large-scale AI models.”
MOREH, considered a major tech company representing the AI semiconductor ecosystem, develops AI cloud solutions for enterprises, including AI infrastructure software. MOREH's solution is fully compatible with various GPU and NPU resources such as AMD as well as the existing NVIDIA, providing AI operators with a wider range of alternatives that are efficient in terms of cost and time.
MOREH was founded in 2020 by CEO Jo Gangwon and other graduates of Seoul National University's Manycore Programming Research Group, which developed Korea’s first Heterogeneous Supercomputer 'Chundoong'. In October last year, it became a hot topic by receiving a Series B round investment worth $22 million from KT, AMD, etc.