September 9, 2026 · Yunus Emre Vurgun
LLM Glossary: 8 Key Terms
Eight terms explain most LLM conversations: tokens, context windows, temperature, RAG, fine-tuning, quantization, KV cache, and emergence. Know these and model cards, papers, and API docs stop being opaque.
The eight terms
| Term | Definition |
|---|---|
| Token | A unit of text processed by LLMs, typically a word fragment or punctuation. |
| Context Window | Maximum number of tokens an LLM can process in a single inference. |
| Temperature | Sampling parameter controlling randomness in generation (0 = deterministic, >1 = creative). |
| RAG | Retrieval-Augmented Generation: combining LLMs with external knowledge retrieval. |
| Fine-tuning | Adapting a pre-trained model to specific tasks with additional training. |
| Quantization | Reducing model precision (e.g., FP16 to INT8) to decrease size and increase speed. |
| KV Cache | Key-Value cache storing attention computations for efficient autoregressive generation. |
| Emergence | Capabilities that appear only at certain model scales, not predictable from smaller models. |
Why tokens and context windows come first
Everything billable about an LLM traces back to tokens: cost per request, maximum input size, and how much history fits. The context window is the budget; every prompt competes for it. That is why this site exists in part — tokens and context windows decide what fits, and compact encodings like TOON fit more of the same facts. Try the difference yourself in TOON Studio.
Fetch it as data
This glossary is the LLM Glossary dataset. Paste it into a system prompt and an agent shares your vocabulary:
curl "https://yjtoon.com/api/dataset/llm-glossary?format=toon"Go deeper with Attention Mechanisms, ML Evaluation Metrics, and reference cards in system prompts.