LG Uplus CTO Sang-yeop Lee (이상엽), executive vice president, left, and OptAIi CEO Jae-ho Lee (이재호) pose for a photo over cooperation on token optimisation. [Photo: LG Uplus]

LG Uplus will develop token optimisation technology needed to operate AI services together with OptAIi, a company specialising in AI model optimisation.

LG Uplus said on Thursday it will jointly research with OptAIi technology that increases computing efficiency of AI models so more requests can be processed with the same resources.

A token is the basic unit of data an AI processes as it understands a user’s question and generates an answer. Token optimisation is a technology that increases the amount of tokens that can be processed with the same graphics processing unit (GPU) resources through model lightweighting and improvements to the computation process.

LG Uplus will be responsible for verifying and applying the technology in a real service environment. OptAIi will handle research and development of technology to run AI models more lightly and quickly.

The GPU-based AI model optimisation research has also secured initial results. The two companies said they improved the response-generation computation process of AI models to fit an actual service environment, increasing the volume of tokens that can be processed on the same GPU by up to four times compared with before.

The two companies previously also cooperated in the on-device AI field. LG Uplus applied lightweighting technology to a small language model (sLM) based on EXAONE so it can run on a smartphone neural processing unit (NPU), an AI-dedicated chip. It cut power consumption by 78 percent and model size by 82 percent while maintaining performance at the same level as the existing CPU method.

The two companies plan to expand the scope of joint research from on-device AI to server GPU environments through this cooperation. They plan to focus on developing technology that improves model computing efficiency so more requests can be processed with the same GPU resources. LG Uplus will gradually apply the secured technology to its AI services and large-scale AI infrastructure operating environments.

LG Uplus CTO Sang-yeop Lee (이상엽) said, "We will expand research on token optimisation technology in cooperation with various partners including OptAIi and produce results that can be applied to actual services."

OptAIi CEO Jae-ho Lee (이재호) said, "We will advance AI optimisation technology that can be verified in an actual service environment with LG Uplus."

Keyword

#LG Uplus #OptAIi #GPU #EXAONE #NPU
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.