PaliGemma 2 Mix - New Instruction Vision Language Models by Google

Google has released PaliGemma 2 mix, a set of vision language models in various sizes and resolutions, fine-tuned for tasks like OCR, captioning, and document understanding. These models can handle general vision-language tasks, text recognition in images, and localization. They accept open-ended prompts and perform well on real-world examples. A demo of the 10B/448 model is available, and the models can be used via the Transformers library for inference and fine-tuning.