Google Cloud announced on the twenty eighth that LG AI Researchers built ‘EXAONE 3.0’ and enterprise AI agent ‘Chat EXAONE’ based on AI-optimized infrastructure and solutions.
On the seventh, LG AI Research Institute introduced the brand new ExaOne 3.0 model. It consists of a lineup of assorted sizes in response to the aim of use, similar to an ultra-light model for on-device AI, a light-weight model for general purposes, and a high-performance model for research. Chat ExaOne, which improves the productivity of corporate members and provides expert-level insights, was also unveiled.
LG AI Research said that it has been collaborating with Google Cloud for the reason that early stages of developing the ExaOne model. In 2021, it introduced Google Cloud TPU to develop and enhance learning of the ExaOne 1.0 model, effectively shortening the time required for AI model learning.
Through the development of ExaOne 3.0, the corporate said that by moreover introducing Google Cloud GPUs, it was possible to enhance compatibility for learning and inference, flexibly transform the architecture, and create an agile AI model learning infrastructure.
Specifically, LG AI Research Center has been specializing in research and development of AI model lightweight technology and price efficiency for the reason that release of ExaOne 1.0. The ‘ExaOne 3.0 7.8B’ model released as open source is the lightweight model with the very best utilization among the many 3.0 models. The ‘ExaOne 3.0 7.8B Instruction Tune’ language model reduced inference processing time by 56%, memory usage by 35%, and operating costs by 72% in comparison with the previous model, ‘ExaOne 2.0’.
ExaOne 3.0 is a bilingual model that may learn and understand Korean and English, and it added that its Korean performance has also recorded the world’s highest level.
The actual ExaOne 3.0 model and the training and operation of ChatExaOne based on it are conducted in Google Cloud. Within the case of ChatExaOne, it have to be connected to the company database held by LG Group affiliates and various external tools, and Google Cloud’s fast and smooth ML Ops pipeline is explained to be suitable.
By way of security, it also said that it has personal information protection, security, and compliance capabilities. Customer data may be protected with data governance and data privacy control functions built into Google Cloud.
LG AI Research is actively reviewing the expansion of using Google Cloud AI infrastructure for continuous model development. The 2 corporations plan to offer ExaOne 3.0 to the Google Cloud AI Marketplace within the near future and expand business cooperation. As well as, they’re exploring ways to develop developer-friendly APIs to reply to the increasing demand for enterprise AI agents and augmented search generation (RAG).
“The rationale we released the Exaone 3.0 7.8B model as open source is to assist AI researchers world wide with their research,” said Lee Hwa-young, managing director of LG AI Research Center. “We were in a position to successfully launch Exaone 3.0 because of the world-class AI infrastructure and customizable architecture provided by Google Cloud.”
Ji Ki-seong, President of Google Cloud Korea, said, “We stay up for continuing our collaboration with LG AI Research, which is leading Korea’s AI core technology, in various fields to open latest horizons in generative AI.”
Reporter Jang Se-min semim99@aitimes.com
