Lesson 1327 of 1596
Mixture of Depths: How AI Models Spend Compute Per Token
Mixture-of-depths lets models skip layers per token to spend compute where it matters; understand it to evaluate efficiency claims honestly.
Creators · AI Foundations · ~20 min read
The premise
Mixture-of-depths trains a router to skip transformer layers on easy tokens, spending compute where input difficulty actually demands it.
What AI does well here
- Reduce average compute per token while preserving downstream quality
- Concentrate compute on tokens with high routing-uncertainty
- Compose with mixture-of-experts for additional efficiency gains
What AI cannot do
- Match dense-model quality on adversarial tail-of-distribution tasks always
- Avoid additional engineering complexity in serving systems
- Replace the need for careful router-training data
Key terms in this lesson
End-of-lesson quiz
Check what stuck
10 questions · Score saves to your progress.
Tutor
Curious about “Mixture of Depths: How AI Models Spend Compute Per Token”?
Ask anything about this lesson. I’ll answer using just what you’re reading — short, friendly, grounded.
Progress saved locally in this browser. Sign in to sync across devices.
Related lessons
Keep going
Creators · 40 min
Mixture-of-Experts: Why MoE Models Behave Differently
Mixture-of-experts architectures route tokens through specialized sub-networks — and the routing creates eval and serving behaviors single-dense models do not have.
Creators · 11 min
Attention deep dive: queries, keys, values, and why it works
Understand attention as a content-addressable lookup over a sequence — and where the analogy breaks.
Creators · 11 min
Tokenization economics: why your bill depends on the tokenizer
Tokenization decisions ripple into cost, latency, and capability — for languages, code, and rare strings.