AI NewsWaveKakao Open-Sources Two Kanana-2 Vision Encoders on Hugging Face
Key points
- Kakao open-sources two vision encoders
- Two models with 426 million and 98 million parameters
- Serve as the eyes of AI that understands images
Kakao has open-sourced two Kanana-2 vision encoders it developed in-house on Hugging Face, an AI development platform. Techworld News, iNews24 and Chosun Biz reported that Kakao announced this on the 21st.
The released models are a medium model with 426 million parameters and a small model with 98 million parameters. Korean outlets spell the small model's name differently: Techworld News and iNews24 write one form, and Chosun Biz another.
Chosun Biz compared a vision encoder to the eyes that help AI understand images. AI Times explained that the release comes as computer use technology, in which AI looks at a screen and operates it directly, draws attention and accurate image reading becomes more important.