Google Releases Gemma-2-JPN 2B AI Model, Announces $150,000 Kaggle Competition

Google has released Gemma-2-JPN, a fine-tuned version of its Gemma 2 2B AI language model optimized for Japanese text. This new model supports Japanese language queries with the same performance as English-only queries on Gemma 2. The model can run locally in browsers using WebGPU and Transformers.js. Additionally, Google has announced a $150,000 Kaggle competition to build Gemma models for various languages. The release includes training materials to help developers adapt Gemma to their own languages. The model's capabilities have been demonstrated in various applications, including code generation and translation. The model is also optimized for mobile and available for demo in the Gradio space shared by Hugging Face.