LG Uplus, a South Korean telecommunications company, said on October 2 that it partnered with OptAI, an AI model optimisation specialist, to develop token optimization technology that cuts graphics processing unit (GPU) use and power use while improving AI processing efficiency.

Tokens are the basic units of data that AI processes when handling user queries and generating responses.

LG Uplus said it will use its AI service operations to test and apply the technology, while OptAI develops lighter, faster AI model execution tools.

In earlier work on a small language model based on LG AI Research’s Exaone model family, the companies said they enabled the model to run on smartphone neural processing units (NPUs).

They said it delivered performance similar to central processing unit (CPU)-based methods, while cutting power use by 78% and reducing model size by 82%.

The companies are now extending the work to server GPU environments. LG Uplus said the technology can raise the number of tokens processed on a single GPU to as much as 4x that of existing methods.

🔗 Source: Chosun Daily

Recent LG Uplus developments