liveBook · Manning

RFiD

As explained in Retrieval Augmented Generation, The Foundational Ideas

RFiD is a Fusion-in-Decoder architecture that guides decoding with scores identifying passages containing answer rationales.

RFiD is a Fusion-in-Decoder architecture for noisy multi-document question answering. When ten retrieved passages contain only three pieces of actual evidence, it learns to identify and prioritize the passages that justify the answer. section 5.6.1

Unlike standard FiD, which may treat related but unsupported passages as useful context, RFiD assigns rationale confidence scores and passes them to the decoder as an explicit signal. A multi-task loss trains answer generation and rationale classification together. section 5.6.1

RFiD is useful when retrieval produces plausible noise alongside supporting evidence:

  • Weights high-rationale passages more heavily during answer generation. section 5.6.1
  • Improves answer faithfulness while exposing which passages the model trusted as supporting evidence. section 5.6.1

How Retrieval Augmented Generation, The Foundational Ideas covers it

How deep it goes
Explains rationale scoring, decoder guidance, and joint training, but does not detail the underlying network internals.
Examples
Uses ten retrieved passages containing three evidential passages and seven related distractors to illustrate rationale-guided trust.
What it assumes
Assumes familiarity with retrieved passages, Fusion-in-Decoder, decoder attention, and answer generation.
What you can do after
Evaluate and design retrieval-augmented answering that prioritizes supporting passages and exposes its evidence preferences.
How it connects
Connects Fusion-in-Decoder with rationale classification, multi-task learning, answer faithfulness, and interpretability.

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.