Fundamentals of Foundation Models

What it is about

You learn what Foundation Models are (e.g., LLMs, CLIP, image generation) and how they are structured, trained and fine-tuned. In the end you understand architecture, scaling, data requirements and distributed training as well as strategies for fine-tuning and alignment/adaptation of models.

What you will be able to do

• Understand the fundamentals of Foundation Models (LLMs, CLIP, image generators)
• Be able to adapt and extend existing Foundation Models
• Apply and adapt optimization methods for training such models
• Master fine-tuning and use of existing models
• Assess data requirements and dependencies of Foundation Models

What the module consists of

• VorlesungIntroduction and technical imparting on architecture, data requirements, scaling and distributed training
• Übungen / Coding-Aufgabenweekly or bi-weekly tasks on theory and implementation; solutions are provided subsequently
• Diskussions-/ÜbungssitzungenClarification of concrete questions on tasks and deepening of individual lecture topics

Teaching method

• Vorlesung mit Folien und TafelConveying the technical foundations and concepts
• Praxisorientierte Coding-BeispieleApplication and implementation of the treated methods
• Regelmäßige Übungsaufgaben mit LösungenSolidification of content through theory and programming tasks
• Diskussions-/ÜbungssitzungenAnswering concrete questions and deepening difficult topics