Google launches Gemma 2, its open-weight model family, in 9B and 27B sizes
The 27B model ran on a single H100 GPU or TPU host, which Google said cut deployment cost while matching models more than twice its size.
Google DeepMindOpen weights & ecosystem