Kat Black

Google launches Gemma 3 270M, a compact 270-million parameter open model designed for task-specific fine-tuning with strong instruction-following capabilities, enabling efficient and cost-effective production systems.

“Co-author of the article announcing Gemma 3 270M model”

Google introduces Gemma 3 270M, a 270-million parameter compact model designed for on-device inference with state-of-the-art instruction-following capabilities and extreme energy efficiency (0.75% battery drain for 25 conversations on Pixel 9 Pro).

“Co-authored blog post on Gemma 3 270M compact model for Google Developers.”

Google introduces TranslateGemma, open translation models available in 4B, 12B, and 27B sizes that outperform larger baselines while supporting 55 languages through knowledge distillation from Gemini.

“Kat Black co-authored the TranslateGemma technical report describing open translation models for 55 language pairs.”

Google DeepMind released FunctionGemma, a specialized open-source version of Gemma 3 270M optimized for function calling and on-device agent capabilities, achieving 85% accuracy on mobile actions after fine-tuning.

“Kat Black co-authored the FunctionGemma release blog post for Google DeepMind.”