Kakao Brain unveils AI development library Trident

Kakao Brain, an artificial intelligence (AI) R&D subsidiary of South Korea's Kakao Corp., announced on Tuesday that it has released a performance library called "Trident" that enhances the training and inference speed of this AI model. The library was made available to developers on GitHub, an open-source platform.

Trident is implemented based on Triton, which is OpenAI's graphics processing unit (GPU) programming language. The library consists of reusable code packages. Trident optimizes the necessary computations for AI model development with optimized kernels. It shares the same usage as PyTorch, a machine learning framework developed by Meta.

According to the company's explanation, AI developers using Trident can reduce the computation time for AI model training and inference by approximately 25% compared to existing methods. Kakao Brain released this library on GitHub to allow developers to utilize it in their AI model research.

"We decided to release Trident to support numerous AI developers in easily creating large AI models and to contribute consistently to the advancement of the AI ecosystem, including the publication of key technologies like large language models and image generation models," CEO of Kakao Brain Kim Il-doo said.

Write to Ju-Hyun Lee at deep@hankyung.com