Google Cloud and Kakao Brain Enhance 'Kallo' Machine Learning Training

[Asia Economy Reporter Seungjin Lee] Google Cloud announced on the 15th that Kakao Brain is enhancing the machine learning training of its AI artist model 'Karlo' using Google Cloud.

Karlo is an AI artist developed by advancing Kakao Brain's previously released ultra-large image generation AI models such as ‘Mindalli’ and ‘RQ-Transformer’. By inputting desired prompts or sentences including subjects, painting styles, locations, times, and colors, the AI can generate a unique image in the world based on the context it understands on its own.

Kakao Brain is utilizing Google Cloud's infrastructure and solutions to train Karlo's machine learning model. Machine learning models like Karlo, which require repeated training on ultra-large multi-modal datasets, demand large-scale computing resources and fast training speeds.

Google Cloud GPUs are the only ones in the industry capable of operating sixteen NVIDIA A100 GPUs on a single node. Kakao Brain reduced bottlenecks by grouping up to sixteen GPU resources per node using cloud GPUs. This minimized performance loss, enhanced computing efficiency, shortened model development time, and consequently reduced costs.

Additionally, the core processes of storing and managing training data on cloud storage and utilizing GPU instances of the Compute Engine for training?essential for Karlo to learn and improve its painting methods?are all conducted on Google Cloud. Kakao Brain has signed a long-term contract with Google Cloud to ensure stable resource supply and improve cost efficiency while carrying out Karlo and other projects.

In November last year, Kakao Brain launched ‘B^ DISCOVER’, Korea’s first AI generation app that draws imagination into pictures using Karlo, to the global market. Earlier this month, the ‘Karlo 1.0’ model was released on GitHub, the world’s largest open-source community. In 2023, Kakao Brain plans to release an improved Karlo model with enhanced quality and features as an open API, further increasing user accessibility.

Jang Hwajin, President of Google Cloud Korea, stated, “Google Cloud supports rapid development of large-scale models and global service expansion based on industry-leading AI and machine learning technologies and powerful infrastructure. We will continue to contribute to the creation and innovation of Korea’s ultra-large AI ecosystem.”