Gemma 2 (9B, 27B)
Gemma 2 in 9B and 27B; Google says the 27B competes with models over twice its size and fits one GPU or TPU host.
Interleaved local-global attention, grouped-query attention, and knowledge distillation for the smaller models (arXiv 2408.00118, 2024-07-31). 27B runs at full precision on one A100 80GB, H100 or TPU host. 2B size added 2024-07-31.
- Date
- Thursday, 27 June 2024
- Lab
- Google DeepMind
- Kind
- open-weights
- Access
- open weights (restricted license)
Sources
- blog.google/technology/developers/google-gemma-2/
- arxiv.org/abs/2408.00118
- developers.googleblog.com/en/smaller-safer-more-transparent-advancing-responsible-ai-with-
This record was checked against its sources on 6 October 2026. How we check