Researcher Shares Lecture Slide on Large Model Pipeline

Michal Valko, an AI researcher at Isara Labs, Inria, and MVA/ENS Paris-Saclay who previously worked on Llama at Meta AI and BYOL/Gemini at DeepMind, posted a tweet. It includes a photo of a lecture slide titled "Large Model Training Pipeline." The slide displays a flowchart beginning with Pretraining from a neural net icon labeled "Random Initialization" to a "Pretrained Model," followed by SFT. Valko wrote: "A lecture on the large-model training pipeline, while the human still does the pointing. Progress, apparently."