JAKARTA – Google has launched a series of open artificial intelligence (AI) models known as Gemma. Developed by Google DeepMind and another team on Google, Gemma is a family of lightweight and sophisticated open models, inspired by Gemini’s previous models. Gemma, which comes from the Latin word meaning “rock gems”, is designed to assist developers and researchers in building AI with responsibility.
Gemma models come in two sizes: Gemma 2B and Gemma 7B, each equipped with a variety that has been pre-trained and adapted to instructions. To ensure the use responsible, Google also provides a new Generatative AI Toolkit. This toolkit provides an important guide and tool for creating a more secure AI application with Gemma.
One key feature of Gemma’s launch is the availability of tools for supervised inference and fine-tuning (SFT) in all major frameworks such as JAX, PyTorch, and TensorFlow via Hard 3.0. In addition, ready-to-use notebooks are also available in Colab and Kaggle, as well as integration with various popular tools such as Hugging Face, MaxText, NVIDIA NeMo, and TensorRT-LLM, which make it easier for developers to start using Gemma.
Gemma models can run directly on laptops, workstations, or Google Clouds with ease of deployment on Vertex AI and Google Kubernetes Engine (GKE). Optimization is carried out on various AI hardware platforms, including NVIDIA GPU and Google Cloud TPU, to ensure leading performance in the industry.
In addition, Google also confirms that Gemma is designed with AI principles on the front lines. This includes the use of automated techniques to filter personal information and other sensitive data from training data sets, as well as extensive use of fine-tuning and reinforcement learning from human feedback (RLHF) to align models tailored to instructions with responsible behavior.
Google also released the Generatative AI Toolkit Responsible together with Gemma to help developers and researchers prioritize building safe and responsible AI applications. This toolkit includes safety classification, modelUGging, and guidance based on the best practices of Google’s experience in developing and implementing large language models.
As part of Google’s commitment to the development community and researchers, Gemma is available for free on Kaggle and Colab. In addition, new Google Cloud users can earn credit of up to 300 US dollars (Rp4.6 million) to start their AI project. Researchers can also apply for Google Cloud credit of up to US $ 500,000 (Rp7.8 billion) to speed up their project.
Through Gemma’s launch, Google hopes to expand Gemma’s family model for various applications. They also promised to host various events and opportunities in the coming weeks to connect, learn and collaborate with Gemma.
With Gemma, Google presents new opportunities for developers and researchers to develop AI responsibly, provide access to advanced technologies and support collaboration across the AI community.
The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language.
(system supported by DigitalSiber.id)