Atlas

Large Language Models

Mixture of Experts

Huge parameter count, small compute per token.

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. L1IntuitionHow to get a model with a trillion parameters that costs about as much to run as one with a tenth of that — by only ever using a slice of it at a time.6m
    40 XP
  2. L2MechanicsThe router, top-k gating, why load balance is a training-stability problem, and what expert parallelism does to a compute cluster.9m
    70 XP
  3. L3CodeBuild a top-k router with a differentiable load-balancing loss in PyTorch, and watch what happens to expert utilization with and without it.11m
    110 XP
  4. L4FoundationsFLOPs accounting for sparse layers, the auxiliary loss objective in full, and the memory-bandwidth reality of running MoE at inference time.10m
    180 XP

33 stars in the atlas.