Pre-training of Smol and Large LLM

Explaining what's new in pre-training: optimization tricks, MoE, stability hacks, and handling long contexts—everything you need to build better LLMs.