Plain-language explanations of terminology from the world of self-hosted LLM inference and serving. Written for someone without an LLM/GPU background — each entry says what the thing is and why it matters in practice.
📖 Read it here: https://fankserver.github.io/llm-glossary/ — organized by category, with full-text search.
- LLM basics
- Model architecture
- How inference works (prefill, decode, caching)
- Speculative decoding
- Quantization
- Fine-tunes & community model names
- Multi-GPU & parallelism
- GPU hardware
- Tokens, templates & parsing
- Embeddings & RAG
- Speech (TTS / ASR)
- Serving engines & runtimes
- Deployment & platform (Docker / Kubernetes)
- Monitoring & benchmarking
- Common vLLM flags
Synonyms and abbreviations are listed in parentheses. Terms link to a primary source (paper, spec, or docs) where one exists.
Missing a term, or found a definition unclear or wrong? Every page on the site has an edit button that takes you straight to its source file — or open an issue or send a PR. Corrections and suggestions of any size are welcome.
Everything here is CC0 / public domain: use it however you like, no attribution needed.