Context Window
Maximum number of tokens a model can attend to in one request.
Обзор
The context window is the number of tokens a model can consider at once, covering the prompt and the response together. It solves the problem that a model has no memory beyond what is in front of it, so it bounds how much source material a request can carry. It exists as a hard architectural limit because attention cost grows with sequence length. It is the constraint that makes retrieval necessary rather than optional.
Граф технологий
Как Context Window связана с остальным стеком.
Использует и зависит от
- Agent Memory technique AI Agents Agent Memory based on Context Window High ✓ source
- Context Engineering technique Prompting Context Engineering based on Context Window High ✓ source
Применяется к
- RAG technique Techniques Context Window used for RAG High ✓ source