More Subsidiary Unveils Small Language Model 'Motif 2.6B'... Accelerates Shift Away from Nvidia

2.6 Billion Parameters, Enhanced Contextual Understanding
Lim Junghwan: "sLLMs Offer Low Power Consumption and High Efficiency"

Motif Technologies (hereafter Motif), a company specializing in developing AMD-based artificial intelligence (AI) models, announced on June 10 that it will release its small language model (sLLM) 'Motif 2.6B' as open source on Hugging Face.

The newly released model is a lightweight, high-performance AI model developed by Motif, based on the efficient use of graphics processing unit (GPU) resources and clustering software (SW) optimization technology provided by its parent company, More, an AI infrastructure specialist. Motif is positioned on the anti-Nvidia front, developing AI models using infrastructure based on AMD GPUs, which compete with Nvidia.

The Motif 2.6B, unveiled on this day, is an sLLM with 2.6 billion parameters. It is reported to outperform not only models in its class but also those with 7 billion parameters.

For accurate performance comparison, Motif applied the same benchmark scores and settings as those disclosed in the technical reports of each developer. As a result, Motif 2.6B demonstrated 134% of the performance of Mistral 7B, which has 7 billion parameters. It achieved particularly high scores in advanced mathematics, science, and coding capabilities.

The most significant technical feature of Motif 2.6B is its enhanced contextual understanding. The model is designed to reduce errors caused by referencing incorrect contexts and generating inaccurate sentences, while focusing on essential core contexts.

Motif CEO Lim Junghwan stated, "Recently, Gartner predicted that enterprise use of small language models will exceed three times that of LLMs by 2027," adding, "Small language models operate on low power and can run without supercomputers. Their high cost efficiency enables a wide range of industrial applications." He continued, "We will further develop Motif 2.6B into on-device AI and agentic AI models that can be used in everyday life."

Going forward, Motif plans to leverage the AI infrastructure software of its parent company, More, along with its own high-performance AI development capabilities, to develop a variety of AI models. By the end of this year, the company aims to release a multimodal model that can convert text into images or videos as open source, contributing to the expansion of the AI industry ecosystem and securing a diverse customer base.