← All notes

Summary

REALM: Retrieval-Augmented Language Model Pre-Training (2020) jointly pretrains a retriever and a language model, so that retrieval is learned as part of pretraining rather than bolted on later. It is an early learned retrieval-augmented model, and a close neighbor of the original RAG paper.

What it introduced: an end-to-end pretraining setup in which the model learns to retrieve documents that help it fill in masked text, with an index that is updated during training.

What it led to: later retrieval-augmented generators, including RAG (a pointer in batch B), Self-RAG, and today's assistant RAG stacks. Live notes: genai-rag-embeddings-vector-search and 2310-11511-self-rag.

Read it alongside the original RAG paper when you want the pretraining-time version of learned retrieval. Where to read it: https://arxiv.org/abs/2002.08909 (v1, 10 Feb 2020).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

@misc{iil_note_6aqq13dk,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{REALM: Retrieval-Augmented Language Model Pre-Training}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2002-08909-realm/},
  note = {Permanent ID iil:note:6aqq13dk. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2002-08909-realm.md.