The release puts part of Kakao's Kanana-2 technology into the hands of outside developers and highlights the industry's growing interest in AI that can operate on personal devices. What a small language model isA parameter is a learned value used by a model when it processes and generates text. Parameter count is not a complete measure of quality, but smaller models generally require less memory and computation. Pruning removes parts of a network under a selected design, while distillation trains a smaller model using signals from a larger teacher model. The company already uses smaller models in features including KakaoTalk message summaries and its AI services.