デプロイメント

Token Budget

定義

The total number of tokens allocated for a model request, encompassing both input (prompt + context) and output. Managing token budgets is central to controlling inference cost in production LLM applications, especially when processing long documents or maintaining conversational history.

関連用語

Context Window

The maximum amount of text (measured in tokens) an LLM can process in a single request, encompassing both the prompt and the generated output. Larger context windows—now exceeding 1 million tokens in some models—enable processing of long documents, codebases, and meeting transcripts in one pass.

Inference Cost

The compute and financial cost of running a model to produce a single prediction or generated response. Inference cost is often the dominant AI operational expenditure at scale and is managed through model compression, caching, quantization, and batching strategies.

Prompt Cache

A technique that caches the key-value (KV) attention states of a repeated prompt prefix, so subsequent requests reuse the pre-computed computation rather than re-running it. Prompt caching reduces latency and cost significantly for applications with long system prompts.

AIの理解にお困りですか？

Physical AI 適合性コールをご予約いただき、これらのAI概念が貴社の業界や課題にどう適用されるかをご相談ください。