Google has announced new versions of its open-source lightweight AI model Gemma 2.
According to the company, the Gemma 2 models come in two sizes of 9 billion and 27 billion parameters, and are more efficient at inference, offering better performance than the first Gemma model.
The company added that the large model, which includes 27 billion parameters, provides performance similar to models that are more than double its size, while the small model, which includes 9 billion parameters, outperforms the Llama 3 8B model and similar open source models, and the parameters are variables in artificial intelligence models. It is used to measure the size and capabilities of models.
The new Gemma 2 model can be hosted on a single NVIDIA A100 80GB Tensor Core or NVIDIA H100 Tensor Core GPU or a single Google TPU AI chip, reducing AI infrastructure costs. The models can also run on NVIDIA RTX or GeForce RTX desktop GPUs via the Hugging Face Transformers library.
Google said that starting next month, Google Cloud customers will be able to use Gemma 2 via the Vertex AI platform, and developers can now try out the new models via Google AI Studio.
Google makes Gemma 2 models available for free, and academic researchers can apply to try them out for academic research.
