2026 MVA Course on Foundations of Large Language Models

This is the official organization for the "Foundations of Large Language Models" offered by the MVA (Master Vision Apprentissage), during the first semester (October -- January) of the academic year 2026 -- 2027. The course is taught in English and is open to all students of the MVA program.

The teachers are David Louapre and Nathanaël Fijalkow.

Please join the Discord server for announcements.

Preliminary plan:

• Session 1: Language models (David, October 1st)
• Session 2: Transformers (Nath, October 29)
• Session 3: Pre-training and scaling laws (David, November 5)
• Session 4: Post-training and alignement (Nath, November 12)
• Session 5: Decoding and inference optimization (David, November 19)
• Session 6: Advanced architectures (Nath, November 26)
• Session 7: Agents (David, December 3)
• Session 8: Advanced topics (Nath, December 10)

A related course is offered for the second semester in 2026 -- 2027 by Marc Lelarge and Guillaume Baudart, called Large Language Models for Code and Proof.