Google Releases Open Source of Small AI Model 'Gemma'

Google has open-sourced 'Gemma,' a small AI model for AI developers and researchers.

On the 21st (local time), Google introduced Gemma through a blog, stating, "It significantly outperforms much larger models on major benchmarks," and "It can run directly on a developer's laptop or desktop computer."

[Image source=Google Korea Blog]

Gemma has 2 billion to 7 billion parameters. Compared to OpenAI's GPT-3.5, which has 175 billion parameters, it is a lightweight model. According to Google, it has the highest performance among models of its class.

Google has open-sourced Gemma. Users can access this model through AI model deployment platforms such as Kaggle, Hugging Face, NVIDIA NeMo, and Google Vertex AI. They can fine-tune the model using their own data to meet the needs of specific applications.