PaliGemma is an open vision-language model designed for tasks such as image captioning, visual question answering, and object detection. Credit: Anna Martyanova/Shutterstock Google has expanded on its Gemma family of AI models, introducing the PaliGemma vision-language model (VLM) and announcing Gemma 2, the next generation of Gemma models based on a new architecture. The company also released the LLM Comparator in open source, an addition to its Responsible Generative AI Toolkit. Google announced the new products on May 14. The company described PaliGemma as a powerful open VLM inspired by the Pali-3 vision-language models, intended to be smaller, faster, and stronger. Built on components from the SigLIP vision model, PaliGemma is designed for a range of vision-language tasks including image and video captioning, visual question answering, understanding text in images, object detection, and object segmentation. PaliGemma can be found on GitHub, Hugging Face, Kaggle, and Vertex AI. Gemma 2, due to be formally launched in coming weeks, features a new architecture designed for “breakthrough performance and efficiency,” Google said. At 27 billion parameters, Gemma 2 offers performance comparable to Llama 3B at less than half the size, Google said. An efficient design reduces deployment expenses, with Gemma 2 fitting on less than half the compute of comparable models. For fine-tuning, Gemma 2 can work with solutions ranging from Google Cloud to tools such as Axolotl. Google also added to its Responsible Generative AI Toolkit by releasing the LLM Comparator in open source. Designed to assist developers with conducting model evaluations, the LLM Comparator is an interactive data visualization tool that allows users to perform side-by-side evaluations of model responses to assess their quality and safety. Generative AIArtificial IntelligenceSoftware DevelopmentTechnology Industry