Kakao Brain releases open-source ViT and ALIGN trained on COYO-700M dataset

AI Impact Summary

Kakao Brain releases open-source ViT and ALIGN trained on the COYO-700M dataset, with demos and training data made public via Hugging Face. The ALIGN-B7-Base and ViT-L/16 variants achieve comparable results to Google’s published baselines on image-text tasks, and the COYO-700M catalog adds metadata that supports finer-grained experimentation. This enables researchers and product teams to reproduce visual-language benchmarks, build cross-modal apps, and prototype solutions without private datasets, while requiring attention to the data’s noisiness and licensing terms for production use.