Google Cloud Collaborates with LG to Enhance Training of ExaOne, the Ultra-Large AI

[Asia Economy Reporter Seungjin Lee] Google Cloud announced on the 22nd that LG AI Research Institute has enhanced the super-large AI model "EXAONE" by adopting Google Cloud's customized machine learning hardware (ASIC) 'Cloud TPU.'

EXAONE is a super-large AI with approximately 300 billion parameters, trained on the world's largest scale of data, representing the highest level of AI in Korea.

Cloud TPU is a processor developed for cloud services to accelerate deep learning, supporting rapid iterative training and execution of cutting-edge machine learning models based on Google Cloud AI and customized high-speed networks.

LG AI Research Institute introduced Cloud TPU and GPUs, which provide industry-leading speed and scalability to EXAONE, effectively reducing the time required for AI model training and establishing a more efficient model training process.

Additionally, LG AI Research Institute is conducting a joint research project with Google Cloud aimed at optimizing and customizing EXAONE's AI training framework, 'Lingvo.' Google Cloud provides outstanding research capabilities and guidelines, especially in natural language processing (NLP) AI model training, minimizing the software and hardware operational burden on LG AI Research Institute while supporting the achievement of SOTA (State-of-the-art), representing global top-tier AI performance.

The two companies plan to further expand their partnership by publicly releasing EXAONE's pre-trained models. They also intend to accelerate the adoption of the EXAONE AI model across all industries, including manufacturing, research, education, and finance.

Hwayoung Lee, Executive Director of LG AI Research Institute, stated, “Thanks to distributed computing performance, we were able to realize the advancement of EXAONE's AI model much faster. We will accelerate research and development to continue LG AI Research Institute's mission to create AI for a better life by building a differentiated super-large AI ecosystem based on EXAONE.”

Sachin Gupta, Vice President of Google Cloud Infrastructure, said, “We are very pleased to provide Google Cloud infrastructure for training Korea's top-level super-large AI model. We will continue to actively support Korean companies and research institutions in discovering future innovative technologies and expanding globally based on industry-leading infrastructure and AI expertise.”

This content was produced with the assistance of AI translation services.