Gemma 2 - Google DeepMind's next-generation open-source artificial intelligence model

What is Gemma 2?

Gemma 2 is a next-generation open-source artificial intelligence model from Google's DeepMind, available in 9 billion and 27 billion parameter versions. Characterized by superior performance, efficient inference speed, and broad hardware compatibility, it rivals models with larger parameter sets, such as Llama 3, Claude 3, and Sonnet. Gemma 2 is designed for easy integration into developer workflows, supports various AI frameworks, and is freely available through platforms such as Google Cloud, Kaggle, and Hugging Face.

Features of Gemma 2

• Parameter sizeGemma 2 currently offers two parameter scales: 9 billion (9B) and 27 billion (27B) parameters, to accommodate different application needs and resource constraints. A 2.6 billion (2.6B) parameter model will be released later.
• Performance optimizationThe 27B version of Gemma 2 is comparable in performance to models with more than twice the number of parameters, demonstrating extremely high performance efficiency. In the LMSYS Chatbot Arena, the 27 billion parameter Gemma 2 command-tuning model outperformed the 70 billion parameter Llama 3 model, and surpassed models such as Nemotron 4 340B, Claude 3 Sonnet, Command R+, and Qwen 72B, ranking first among all open-source weighted models.
  
• Reasoning efficiencyGemma 2 is specifically optimized for inference, enabling it to run at full precision on a single high-end GPU or TPU without requiring additional hardware resources, thus significantly reducing the cost of use.
• Hardware compatibilityGemma 2 can run quickly on a variety of hardware platforms, including personal computers, workstations, gaming laptops, and cloud servers.
• Open licenseGemma 2 uses a business-friendly license agreement, allowing developers and researchers to freely share, use, and commercialize their applications.
• Framework supportGemma 2 is compatible with several mainstream AI frameworks, including Hugging Face Transformers, JAX, PyTorch, and TensorFlow, allowing developers to choose the appropriate tools according to their preferences.
• Deployment toolsGoogle offers the Gemma Cookbook, a resource library containing practical examples and guides to help users build applications and fine-tune Gemma 2 models.
• Responsible AIGoogle offers a range of tools and resources, such as the Responsible Generative AI Toolkit and LLM Comparator, to support developers and researchers in building and deploying AI responsibly.

How to use Gemma 2

Gemma 2 can be easily used with commonly used tools and workflows, and...Hugging Face It is compatible with mainstream AI frameworks such as Transformers, JAX, PyTorch, and TensorFlow, and can be used with native Keras 3.0, vLLM, and others.Gemma.cpp,Llama.cppandOllamaImplementation. Furthermore, Gemma also...NVIDIA TensorRT-LLMOptimized to run on NVIDIA accelerated infrastructure or as a...NVIDIA NIMInference microservices run and will targetNVIDIA's NeMoOptimize.

Gemma 2 is now availableGoogle AI StudioLaunched in [location], users can test its full performance at 27B without any hardware requirements. Developers can also [access/test] from [location].KaggleandHugging Face ModelsDownload the model weights for Gemma 2.Vertex AI Model Gardencoming soon.

For ease of research and development, Gemma 2 can also be used via...KaggleAlternatively, you can use a Colab notebook for free. First-time Google Cloud customers are eligible to receive [this benefit].$300 credit limitAcademic researchers can apply.Gemma 2 Academic Research ProgramThey wanted to obtain Google Cloud credits to accelerate their research using Gemma 2.ApplyThe opening hours are from now until August 9th.