TIL: Gemmaverse
Strictly speaking, last night I learned about the compendium of open, smaller models made available by Google. Thanks to the exceedingly generous Hugo Bowne-Anderson (do check out the Vanishing Gradients podcast) for hosting a pop-up seminar with Ravin Kumar of Google DeepMind.
Gemma Models Overview
Gemma is a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. Developed by Google DeepMind and other teams across Google, Gemma is named after the Latin gemma, meaning precious stone. The Gemma model weights are supported by developer tools that promote innovation, collaboration, and the responsible use of artificial intelligence (AI). You can get multiple variations of Gemma for general and specific use cases:
Gemma 3: Solve a wide variety of generative AI tasks with text and image input, support for over 140 languages, and long 128K context window.
CodeGemma: Complete programming tasks with this lightweight, coding-focused generative model.
PaliGemma 2: Build visual data processing AI solutions with a model that’s built to be fine-tuned for your image data processing applications and available in multiple resolutions.
ShieldGemma 2: Evaluate the safety of generative AI models’ input and output against defined policies.
Many more Gemma variants are available from Google and our AI developer community. Check them out on Kaggle Models and Hugging Face. Get inspired by what our community members have built with Gemma in the Gemmaverse.
I’d heard of a few of the Gemma models but didn’t realize how diverse the collection had become.
Bonus: Kumar dropped a Colab Notebook on finetuning the FunctionGemma 270M model.
Built using