The foundation models currently attracting major attention are large, potentially multimodal, universally pre-trained AI systems that provide the basis for rapidly deploying AI in specific applications. Among them is the Fraunhofer language model Teuken-7B, trained in 24 official European languages with exceptional resource efficiency. Through our work on Teuken, Fraunhofer has built the capabilities needed to process large volumes of data and to develop specialized foundation models. In doing so, we help safeguard the digital sovereignty of Germany and Europe.
Task-oriented AI refers to technologies developed to perform specific tasks. This includes both foundation models and foundational methods such as machine learning, deep neural networks, and forecasting or optimization techniques. These approaches are designed to deliver optimal solutions for our customers. Whether in data selection, the configuration of training procedures, or algorithm and hardware development, we conduct research on resource-efficient, cost-effective, low-energy, low-maintenance, time- and data-efficient systems with low latency for real-world domain applications.