Templates reduce drift. They let editors expand dense LLM topics without turning every page into a different shape or burying limitations below the fold.
Universal page skeleton
Title
Lead: define the topic and why it matters.
Status: quality class, review date, source state.
Background: historical or technical context.
Mechanics: how the thing works.
Applications: where it is used.
Limitations: failure modes, caveats, unresolved questions.
Examples: compact, source-backed examples.
Related pages: internal links.
References: primary or reliable sources.Template variants
Core concept
Use for Transformer, self-attention, tokenization, hallucination, bias, and privacy. Required sections: definition, mechanics, examples, limitations, references.
Model family
Use for GPT, LLaMA, BERT, and related model pages. Required sections: provider, release status, architecture, capabilities, limits, source dates.
Training method
Use for pretraining, fine-tuning, RLHF, adapters, and prompt tuning. Required sections: objective, data, workflow, tradeoffs, common mistakes.
Application
Use for chatbots, summarization, code generation, and document workflows. Required sections: task shape, architecture, examples, metrics, safety notes.
Tool or infrastructure
Use for Hugging Face, LangChain, RAG, vector databases, and local model stacks. Required sections: purpose, setup, integration points, risks, alternatives.
Evaluation page
Use for metrics, benchmarks, and leaderboards. Required sections: what it measures, data source, scoring method, limitations, contamination risk.
Minimum publish checklist
- The lead answers what the topic is, why it matters, and who uses it.
- The body separates mechanics, examples, applications, and limitations.
- The page includes at least one internal link to a related handbook route.
- The page identifies primary sources where available and dates time-sensitive claims.
- The status panel shows quality class, authority boundary, and last review date.