Gemma's open-source release contrasts with Gemini's release, which is a closed AI model that developers can build on through APIs or Google’s Vertex AI platform. By making Gemma open source, Google aims to encourage more experimentation with its AI. Both models will be available with a commercial license, but Google often restricts its models from being used for certain tasks like weapons development. Gemma will also come with “responsible AI toolkits” to help developers manage the risks associated with open models.
Key takeaways:
- Google has released Gemma 2B and 7B, a pair of open-source AI models that are lighter and faster than its flagship Gemini, and are suitable for smaller tasks like simple chatbots or summarizations.
- Gemma models will be available via Kaggle, Hugging Face, Nvidia’s NeMo, and Google’s Vertex AI, and can be used for free in Kaggle by developers.
- Unlike Gemini, which is a closed AI model, Gemma is open-source, allowing more people to experiment with Google's AI. However, it will also ship with "responsible AI toolkits" to help manage the risks associated with open models.
- Google is also offering commercial licenses for both model sizes, and researchers can apply for up to $500,000 in cloud credits to use the models.