Developers: | |
Date of the premiere of the system: | February 2024 |
Branches: | Information Technology |
Technology: | Application Development Tools |
2024: Product Announcement
On February 21, 2024, Google announced open source artificial intelligence models Gemma 2B and Gemma 7B, which can be used by everyone. It is possible to solve such problems as document analysis, creation of chat bots, etc.
It is noted that Gemma is a family of light AI models created on the basis of the same research and technologies that formed the basis of closed Gemini solutions. Specialists from Google DeepMind and other Google divisions took part in the development. You can run Gemma on a personal computer, workstation, or in the cloud.
The Gemma 2B and Gemma 7B models have 2 billion and 7 billion parameters, respectively. Google provides third-party developers with all the necessary tools with which you can additionally train models and adapt them to specific projects. Integration with Hugging Face, MaxText, NVIDIA NeMo and TensorRT-LLM has been implemented. The models were optimized to work on hardware platforms with graphics processorsNvidia and proprietary Google Cloud TPU chips.
Responsible commercial use of Gemma in organizations of any size is allowed. To make Gemma's pre-trained models safe and reliable, Google used automated methods to exclude certain information from training arrays that could be sensitive or personal. At the same time, the method of training with reinforcement based on human feedback (RLHF) was used. Together with Gemma, a specialized set of Responsible Generative AI Toolkit for researchers and developers is presented. It is noted that compared to more complex large language models, solutions of the Gemma family provide increased speed.[1]