liveBook · Manning

Caching

As explained in Retrieval Augmented Generation, The Foundational Ideas

Caching stores reusable computation results or retrieved information to avoid repeating expensive operations.

Caching reuses evaluation work instead of repeating expensive operations. In RAG evaluation, cached claim extractions and verifications are especially useful when the same context–answer pair is tested against multiple retrieval configurations. section 14.5.3

Practical caching patterns include:

  • Cache repeated evaluation calls and group independent calls into batches to reduce API overhead. section 14.5.3
  • Persist evaluation results in a database or versioned JSON so score trends can be tracked without re-evaluating historical samples. section 14.5.3
  • Treat caching as an optimization that strengthens RAG rather than replacing retrieval or context engineering. section 1.3.4

Repeated retrieval experiments

When several retrieval configurations produce the same context–answer pair, reuse its cached claim extraction and verification results instead of rerunning those steps. section 14.5.3

How Retrieval Augmented Generation, The Foundational Ideas covers it

How deep it goes
Explains result reuse and structured persistence, but does not detail eviction, invalidation, or cache implementation.
Examples
Uses repeated context–answer pairs and cached evaluation results stored in databases or versioned JSON.
What you can do after
Operate/configure caching to skip redundant extraction steps and retain evaluation results for trend tracking.
How it connects
Connects caching with RAG, evaluation pipelines, batching, retrieval configurations, and long-context models.

Also explained in

Compare how 49 books explain caching →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Retrieval Augmented Generation, The Foundational Ideas. The section links point to the parts of the book it draws on.