Building your own LLM

Sapienza School for Advanced Studies (SSAS)

Timetable: Dec. 2-3, 9-10, 16-17, 4-7 PM (Tuesday), 5-7 PM (Wednesday).
Attendance: Aula G0, Palazzina G, Viale Regina Elena 295

Overview

The course explores the use of large language models (and more generally, large multimodal models), with a focus on their application in scientific research. It begins with a technical introduction, describing how the field has evolved in recent years (e.g., with the emergence of so-called “reasoning models”). The second, more practical part explains how to build specialized models for specific tasks - for instance, by adding new tools or modalities (e.g., audio) to existing models. The third part focuses on understanding and steering the models via recent work on mechanistic interpretability.

Prerequisites: The first part presents AI models both from a general perspective and by delving into their mathematical formulation, thus requiring a background in linear algebra. The second part requires knowledge of Python. Some parts of the course can still be followed without these prerequisites—in that case, it is recommended to contact the instructor directly.

Material

	Material
Slides	Link
Labs	Notion page

Advanced reading material

• TBD