Google Unveils Open LLM 'Gemma'

Google has released Gemma, a new open LLM family based on Gemini models.

The models are available in two sizes, 2B and 7B parameters, each offered as a Base model and an Instruction-tuned model. All models support an 8K context length and can run efficiently on consumer GPUs and TPUs.

Key features and updates are as follows:

• Performance: Gemma 7B performs on par with existing 7B-class open models such as Mistral 7B, while Gemma 2B is optimized for on-device and CPU environments.
• Version update: One month after launch, an additional Gemma 1.1 version was released with improved coding ability, factuality, instruction-following, and multi-turn quality.
• Ecosystem support: Through Hugging Face, model cards, Google Cloud integration, Inference Endpoints support, and fine-tuning examples using 🤗 TRL are provided.