Google announces Gemma 2, the next generation of its open, lightweight AI models, built for improved performance and efficiency. There are two sizes available, 9 billion (9B) and 27 billion (27B) parameters, with the Model 27B offering performance comparable to larger models while the Model 9B outperforms other models of similar size. A smaller model with 2.6 billion parameters is also in the works for smartphone AI scenarios. Gemma 2 can be hosted on various GPUs, reducing infrastructure costs, and will be compatible with Google Cloud’s Vertex AI. Developers can test the model in Google AI Studio. During training, Google conducted extensive testing and evaluation to ensure security and mitigate biases. Gemma 2 is available for free through Kaggle and Colab, with an Academic Research Program for researchers. This commitment to open access and responsible AI development sets a positive example for the industry.
Article Source
https://www.neowin.net/amp/googles-gemma-2-delivers-powerhouse-performance-at-a-fraction-of-the-cost/
