← back
Unveiling the latest Gemma model advancements: Kathleen Kenealy
Takeaway
Gemma 2 (9B, 27B) ships as a multi-framework, safety-tuned open model competitive with 2-3x larger LLaMA/Grok models, expanding a family that already includes code, recurrent and vision-language variants.
Summary
- Kathleen Kenealy (Google DeepMind, Gemma tech lead) walks through the Gemma open-model family built on the same research stack as Gemini.
- Variants in the family: Gemma 1.0/1.1 base LLMs, CodeGemma (code-tuned), RecurrentGemma (state-space architecture for fast long-context inference, in 2B and 9B sizes), and PaliGemma (SigLIP vision encoder + Gemma decoder for VQA, captioning, detection, segmentation).
- Launches Gemma 2 at 9B and 27B during the talk — both claim best-in-class for their size and competitive with models 2-3x larger; 27B is in the ballpark of LLaMA 3 70B and outperforms Grok on several benchmarks.
- Safety-by-design: manual data-set inspection, safety evals from earliest ablations through final state-of-the-art Gemini-grade safety evals — emphasizing trustworthy behaviour regardless of fine-tuning.
- Broad framework compatibility: TensorFlow, JAX, Keras, PyTorch, Ollama, Transformers; 27B is hosted in Google AI Studio for instant try-out.
gemmaopen-weightsgoogle
Original description
Lets cut through the buzzwords and get to the point. In this talk we'll unveil the abilities of Gemma and Gemini models, and discuss how to design and build killer apps with LLMs. We'll cover the essential tools, techniques, and best practices you need to know, without the fluff. We'll explore how developers can use Google Cloud's Vertex AI platform to build safe, secure applications with open models. Using Google's Gemma model, we'll dive into an end-to-end example of model discovery, tuning, and deployment as part of real-world application. Recorded live in San Francisco at the AI Engineer World's Fair. See the full schedule of talks at https://www.ai.engineer/worldsfair/2024/schedule & join us at the AI Engineer World's Fair in 2025! Get your tickets today at https://ai.engineer/2025