AI Spring School 2026

April 4-5, 2026

Syed Babar Ali School of Science and Engineering
Lahore University of Management Sciences

Overview

The AI Spring School focuses on the foundational architectures and cutting-edge frontiers of Artificial Intelligence, bringing together students, researchers, and professionals. Over the course of two days, we will explore the core mechanisms driving modern AI, with a specific emphasis on Large Language Models (LLMs), generative systems, and agentic AI. For full details regarding the event, registration, and broader scope, please visit the official AISS 2026 webpage.

Workshop: Language Modeling from Scratch

As part of the Spring School, this technical workshop is divided into interactive sessions. The presentation slides for these sessions are embedded below.

• Session 1: Why LLMs behave the way they do

  • Topics: Inference vs. training; Next-token prediction; Loss, gradients, updates.

  • Full-screen view: Click here to open in a new tab

• Session 2: How a Transformer Computes the Next Token

  • Overview: A granular look inside the Transformer architecture. This session breaks down tokenization, embeddings, the mathematical mechanics of multi-head self-attention (Query, Key, Value matrices), positional encoding, and the complete forward pass of a GPT block.

  • Full-screen view: Click here to open in a new tab

Acknowledgements

Special thanks to Omer Abdul Jalil for his invaluable help in developing and preparing the interactive presentation slides used in these workshop sessions.