Level 5: AI / LLM systems · L5.3 LLM inference serving & GPU capacity
Quantization, speculative decoding, prefix caching
Choose optimizations and know their quality risks.
13 min · Pro
What you take away
- All three attack the same bottleneck differently: read fewer bytes, spend fewer sequential passes, or skip work already done.
And 3 more inside the class, in Pro.
How the class is laid out
- The problem
- Three ways round the same wall
- Choosing, and the honest caveat
- Check your understanding