Google releases Gemma, open models based on Gemini
Google released Gemma, open LLMs in 2B and 7B sizes (base and instruction-tuned) with an 8K context, built from Gemini research, with day-one Hugging Face support.
What happened
Google released Gemma, a family of open LLMs. Hugging Face announced the integration: models on the Hub, Transformers v4.38, Google Cloud and Inference Endpoints, and a single-GPU fine-tuning example with TRL.
The technical details
From the Hugging Face post:
- two sizes: 7B (efficient on consumer GPUs and TPUs) and 2B (CPU and on-device)
- each with base and instruction-tuned versions: four models in total
- 8K-token context
- based on technology from Gemini
- all variants can run on consumer hardware without quantisation
Why it matters
A major lab releasing open weights in small sizes was a big deal for local and on-device use. Gemma later grew into several generations (Gemma 2, 3, 3n and 4 appear in later posts here).
Sources
- Welcome Gemma - Google's new open LLM — Hugging Face
- Transformers v4.38.0 release notes — GitHub · huggingface/transformers
This is a summary written from the linked public sources. Figures and numbers are as reported there.