Josh Woodward, Google’s VP of Google Labs, stated that the Gemma models have been downloaded millions of times across various services. He emphasized that the new 27-billion model has been optimized to run on Nvidia’s next-gen GPUs, a single Google Cloud TPU host, and the managed Vertex AI service. While performance data for Gemma 2 is yet to be shared, Woodward noted that it's already outperforming models twice its size.
Key takeaways:
- Google announced new additions to Gemma, its family of open models, at the Google I/O 2024 developer conference, including the launch of Gemma 2, a 27 billion parameter model.
- PaliGemma, a pre-trained Gemma variant for image captioning, image labeling and visual Q&A use cases, is already available.
- Josh Woodward, Google’s VP of Google Labs, noted that the Gemma models have been downloaded more than “millions of times” and the new 27-billion model is optimized to run on Nvidia’s next-gen GPUs, a single Google Cloud TPU host and the managed Vertex AI service.
- Google claims that Gemma 2 is already outperforming models two times bigger than it, but more data and developer feedback is needed to confirm its performance.