Review — RAG Model: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

Recent Langchain Framework Also Based on RAG Model for Dense Text Retrieval

Sik-Ho Tsang
6 min readSep 22, 2023

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Retrieval-Augmented Generation (RAG)
, by Facebook AI Research; University College London; New York University
2020 NeurIPS, Over 870 Citations (Sik-Ho Tsang @ Medium)

Dense Text Retrieval
==== My Other Paper Readings Are Also Over Here ====

  • A general-purpose fine-tuning recipe is proposed for retrieval-augmented generation (RAG) — models which combine pre-trained parametric and non-parametric memory for language generation.
  • RAG models where the parametric memory is a pre-trained seq2seq model and the non-parametric memory is a dense vector index of Wikipedia, accessed with a pre-trained neural retriever.


  1. Retrieval-Augmented Generation (RAG) Conceptual Idea
  2. RAG Model
  3. Results

1. Retrieval-Augmented Generation (RAG) Conceptual Idea

There are 2 RAG model variants: RAG-Sequence Model and RAG-Token Model.

1.1. RAG-Sequence Model

  • The RAG-Sequence model uses the same retrieved document to generate the complete sequence.

Technically, it treats the retrieved document as a single latent variable that is marginalized to get the seq2seq probability p(y|x) via a top-K approximation.

  • Concretely, the top K documents are retrieved using the retriever, and the generator produces the output sequence probability for each document, which are then marginalized:

1.2. RAG-Token Model

  • In the RAG-Token model, a different latent document is drawn for each target token and marginalized accordingly. This allows the generator to choose content from several documents when producing an answer.

Concretely, the top K documents are retrieved using the retriever, and then the generator produces a distribution for the next output token for each document, before marginalizing, and repeating the process with the following output token, Formally:

1.3. Sequence Classification Task

  • RAG can be used for sequence classification tasks by considering the target class as a target sequence of length one, in which case RAG-Sequence and RAG-Token are equivalent.

2. RAG Model

RAG Framework Overview

2.1. Retriever: DPR

  • The retriever is based on DPR [26]. DPR follows a bi-encoder architecture:
  • where d(z) is a dense representation of a document produced by a BERT_BASE document encoder, and q(x) is a query representation produced by a query encoder, also based on BERT_BASE.
  • Calculating top-k((.|x)), the list of k documents z with highest prior probability (z|x), is a Maximum Inner Product Search (MIPS) problem, which can be approximately solved in sub-linear time [23], i.e. using FAISS [23].

A pre-trained bi-encoder from DPR is used to initialize the retriever and to build the document index. This retriever was trained to retrieve documents which contain answers to TriviaQA [24] questions and Natural Questions [29].

The document index is referred as the non-parametric memory.

2.2. Generator: BART

The generator component (yi|x, z, y1:i-1) could be modelled using any encoder-decoder. BART-large, a pre-trained seq2seq Transformer with 400M parameters, is used.

  • BART was pre-trained using a denoising objective and a variety of different noising functions, outperforms comparably-sized T5.
  • To combine the input x with the retrieved content z when generating from BART, they are simply concatenated.

The BART generator parameters θ is refered as the parametric memory.

2.3. Training

  • The retriever and generator components can be trained jointly.
  • Given a fine-tuning training corpus of input/output pairs (xj, yj), the RAG model can be minimized using the negative marginal log-likelihood:

Yet, updating the document encoder BERT_d during training is costly. Thus, The document encoder (and index) is keeping fixed, only the query encoder BERT_q and the BART generator are fine-tuned.

2.4. Decoding

  • For RAG-Token model, it can be seen as a standard, autoregressive seq2seq generator with transition probability:
  • To decode, p’θ(yi|x, y1:i-1) is plugged into a standard beam decoder.
  • For RAG-Sequence, beam search is run for each document z, scoring each hypothesis using:
  • This yields a set of hypotheses Y, some of which may not have appeared in the beams of all documents.
  • There are two decoding methods:
  • Thorough Decoding: To estimate the probability of an hypothesis y, an additional forward pass is run for each document z for which y does not appear in the beam, multiply generator probability with (z|x) and then sum the probabilities across beams for the marginals.
  • Fast Decoding: For longer output sequences, |Y| can become large, requiring many forward passes. For more efficient decoding, a further approximation can be made: (y|x, zi)≈0 where y was not generated during beam search from x, zi.

3. Results

  • A single Wikipedia December 2018 dump is used for the non-parametric knowledge source. Each Wikipedia article is split into disjoint 100-word chunks, to make a total of 21M documents.
  • The document encoder is used to compute an embedding for each document, and build a single MIPS index using FAISS [23] with a Hierarchical Navigable Small World approximation for fast retrieval [37].
  • During training, the top k documents are retrieved for each query. k∈{5,10} is considered for training and k is set for test time using dev data.

3.1. Open-domain Question Answering

Table 1: Unlike REALM and T5+SSM, RAG enjoys strong results without expensive, specialized “salient span masking” pre-training [20].

3.2. Abstractive Question Answering

Table 2: RAG-Sequence outperforms BART on Open MS-MARCO NLG by 2.6 Bleu points and 2.6 Rouge-L points.

  • Table 3: Some examples are shown above.

3.3. Jeopardy Question Generation

Table 2: RAG-Token performs better than RAG-Sequence on Jeopardy question generation, with both models outperforming BART on Q-BLEU-1.

Table 4: Evaluators indicated that BART was more factual than RAG in only 7.1% of cases, while RAG was more factual in 42.7% of cases, and both RAG and BART were factual in a further 17% of cases.

  • Figure 2: A visualization is shown above.

3.4. Fact Verification

Table 2: For 3-way classification, RAG scores are within 4.3% of state-of-the-art models, which are complex pipeline systems.

3.5. Further Study

Table 5: RAG-Sequence’s generations are more diverse than RAG-Token’s, and both are significantly more diverse than BART without needing any diversity-promoting decoding.

Table 6: Learned retrieval improves results for all tasks.

Figure 3 (left) shows that retrieving more documents at test time monotonically improves Open-domain QA results for RAG-Sequence, but performance peaks for RAG-Token at 10 retrieved documents.

Figure 3 (right) shows that retrieving more documents leads to higher Rouge-L for RAG-Token at the expense of Bleu-1, but the effect is less pronounced for RAG-Sequence.



Sik-Ho Tsang

PhD, Researcher. I share what I learn. :) Linktree: for Twitter, LinkedIn, etc.