Atlas

Inference Optimization

Speculative Decoding

A small model guesses, a big model checks. Free speed.

You can jump straight in, but this star assumes The KV Cache. 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 a small, fast model can speed up a big, slow model's generation without changing its outputs.5m
    40 XP
  2. L2MechanicsThe accept/reject mechanics that make speculative decoding exactly lossless.8m
    70 XP
  3. L3CodeImplement the speculative decoding accept/reject loop in PyTorch-style pseudocode.12m
    110 XP
  4. L4FoundationsProving speculative decoding is exactly lossless, and modeling its expected speedup.15m
    180 XP

33 stars in the atlas.