// ⚡ Inference Acceleration & Serving

Temperature & Nucleus (Top-p) Sampling

Controls randomness by sampling only from the top-p probability mass.

温度与 Top-p 采样 2019 Beginner core

IN PLAIN WORDS

Greedy decoding repeats; pure sampling rambles. Top-p keeps the smallest set of tokens covering probability p, and temperature sharpens or flattens the distribution.

Used in: temperature / top_p in every LLM API

TIMELINE · prerequisites → this → what builds on it

2017 Transformer &…2019 Temperature & Nuc…2022 Speculative D…2023 LLM Watermark…

01Original paper

The Curious Case of Neural Text Degeneration

Ari Holtzman, Jan Buys, Li Du, et al. · 2019

arXiv ↗ ✓ Verified · arXiv API

02Where it sits

Learn the left column first; the right column builds on this one.

Learn first

Transformer & Self-Attention

Temperature & Nucleus (Top-p) Sampling

Then you can learn

Left: prerequisites · right: builds on this · dashed: related. Full map →

03Learning resources

Official docs, explainers, courses and reference code. All links checked.

Courses

huggingface.co ↗MULTI

Explainers

Lilian Weng：Large Transformer Model Inference Optimization ↗EN

04Where the course teaches it

U02 Hallucination, Reliability & Limits Why models confidently make things up, and how to check and reduce the risk. 2 h2 algorithms1 papers U60 Decoding, Sampling & KV Cache Prefill and decode, temperature/top-p, KV cache and the memory-bandwidth bottleneck. 6 h4 algorithms2 paperslesson ready

On learning paths:Understanding an LLM End to End

For AI / data

Verification: verified, checked 2026-10-21.
JSON: /api/learn/algorithms.json (id: sampling)