Topics
Recall that in vanilla RAG, we chunk docs and a subset of the chunks (top K ranked by relevance to the input question) is passed to the LLM as context. One of the metrics for evaluating this retrieval component is 1-recall@K aka chunk retrieval failure rate.
Measures the percentage of relevant documents that fail to be retrieved within the top K chunks