LG U+ and OptAI collaborate to develop AI token optimization technology to reduce GPU and power usage

Reduce GPU/power usage and improve AI service processing efficiency

(Seoul = Yonhap News) Reporter Park Hyeong-bin = LG U+ [032640] announced on the 2nd that it will work with Opt AI, a company specializing in AI model optimization, to develop token optimization technology that reduces GPU and power usage required for AI operation and increases processing efficiency.

LG U+ is responsible for technology verification and application based on its AI service operation experience, and Opt AI researches and develops lightweight technology that reduces the amount of calculation and size of AI models.

Previously, both companies applied lightweight technology to the 'EXAONE'-based small language model (sLM) to enable it to run on NPU, a smartphone dedicated to AI semiconductor. While maintaining similar performance to the existing CPU method, power consumption was reduced by 78% and model size by 82%.

Following on-device AI, the two companies are expanding the scope of their joint research to the server GPU environment. In particular, we have secured technology to increase the amount of tokens that can be processed on the same GPU by up to four times the existing amount by optimizing the calculation method in the process of generating answers by the AI ​​model.

On the 1st, the two companies jointly held a seminar to share research trends and industrial application cases of AI operation efficiency technology and discussed ways to apply related technologies.