Google Launches Gemma, Tops Llama-2

Google Launches Gemma, Surpassing Llama 2.0, February 22

Google has joined the battleground of open-source large models today by announcing the launch of a new generation open model called Gemma, aimed at helping developers and researchers to responsibly build artificial intelligence. Google claims that Gemma outperforms Llama-2.

Gemma is a series of lightweight, cutting-edge open models built using the same research and technology that created the Gemini model. Developed by Google DeepMind and other teams at Google, Gemma was inspired by Gemini, with its name reflecting the Latin word “gemma,” meaning “gem.”

In addition to the model weights from Google, tools have been released to support developers in innovation, foster collaboration, and guide the responsible use of Gemma models.

Google has released two sizes of model weights: Gemma 2B and Gemma 7B. Variants trained and fine-tuned through pretraining are released for each size.

The new Responsible Generative AI toolkit provides guidance and foundational tools for creating safer AI applications using Gemma.

Google offers inference and supervised fine-tuning (SFT) toolchains across all major frameworks through native Keras 3.0: JAX, PyTorch, and TensorFlow.

Ready-to-use Colab and Kaggle notebooks, along with integrations with popular tools such as Hugging Face, MaxText, NVIDIA NeMo, and TensorRT-LLM, allow users to easily get started with Gemma.

Pretrained and fine-tuned Gemma models can run on your laptops, workstations, or Google Cloud, and can be easily deployed on Vertex AI and Google Kubernetes Engine (GKE).

Optimizations across multiple AI hardware platforms ensure industry-leading performance, including NVIDIA GPUs and Google Cloud TPUs.

The terms of use allow all organizations (regardless of size) to responsibly utilize and distribute Gemma for commercial purposes.

Google states that Gemini is currently the largest and most powerful AI model widely used today. Compared to other open models, this enables Gemma 2B and 7B to achieve top performance within their size ranges. Gemma models can run directly on developers’ laptops or desktop computers.

Leidi was founded by media personality Lei Jianping. For republishing, please attribute the source.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.