Atlas

Large Language Models

Pretraining

Next-token prediction over a large slice of the internet.

You can jump straight in, but this star assumes The Transformer. Not recommended as a first stop.

Four layers of depth

Each layer ends with a quiz. Finish layer 4 and you own this concept.

  1. L1IntuitionThe single objective — predict the next token — that, scaled up enough, produces everything an LLM can do.7m
    40 XP
  2. L2MechanicsThe training loop mechanically: loss, batching, learning rate schedules, checkpointing, and how a run is actually managed for months.11m
    70 XP
  3. L3CodeA real (small-scale but structurally complete) pretraining loop: AdamW, warmup+cosine schedule, gradient accumulation, checkpointing.15m
    110 XP
  4. L4FoundationsCross-entropy and perplexity derived properly, the Chinchilla compute-optimal arithmetic, and where the FLOPs actually go on real hardware.15m
    180 XP

Where this leads

33 stars in the atlas.