MIT and IBM released ChartNet, a 1.7-million-sample synthetic training dataset that lets compact open-source vision-language ...
Google DeepMind just rolled out Gemma 4 12B, a 12-billion-parameter model that can parse text, images, audio, and video ...