AI Systems · core
Sampling & Decoding
Temperature, top-k, greedy decoding.
Mental model
After the model outputs a probability for each next token, you have to pick one. Greedy picks the top — safe but dull. Temperature, top-k, and top-p (nucleus) introduce controlled randomness. Pick by how much variety the task can tolerate.
How to study Sampling & Decoding
Begin by restating the mental model in your own words, then connect it to a concrete system you have built or operated. Name the mechanism, the constraint it addresses, and the trade-off it introduces. Use Stanford CS336 — L10 Inference (Spring 2025), The Curious Case of Neural Text Degeneration (Holtzman et al.), The Illustrated GPT-2 — output sampling to check details, but close the source before writing your explanation. Retrieval is the learning step; rereading is only preparation.
Next, compare Sampling & Decoding with Model Evaluation. Ask what changes in correctness, latency, resource use, operability, and failure recovery. Complete Temperature scaling logits and preserve the command, input, output, and one failed attempt as evidence. Finish by explaining the idea without jargon to someone who has not studied the track.
Proof of understanding
- Explain the mechanism from first principles and identify the state it reads or changes.
- Give one situation where the concept is the right choice and one where it is not.
- Predict a realistic failure mode before running the drill, then compare the prediction with evidence.
- Connect the result to a roadmap or build artifact instead of treating the concept as isolated trivia.
Learn from primary sources
Practice and explain it back
Temperature scaling logits
Logits [2,1,0]. Apply temperature T=2 and T=0.5 before softmax. Which T makes the distribution sharper?
Expected evidence: T<1 sharpens (more greedy); T>1 flattens.
Open the interactive drill →Review prompts
- Top-k and top-p both truncate the distribution. When does the difference matter?
Build evidence
Synthesize: Inference & Serving
Build a production mental model for inference engines, memory, kernels, routing, hardware, and serving economics. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks