Scholar Feed

Retrieval-augmented generation

xRAG

xRAG: Extreme Context Compression for Retrieval-augmented Generation with One Token

Superseded baseline#58 of 1,179 most-superseded · first seen May 22, 2024

Superseded — cited as a baseline and beaten by newer methods

3 papers critique it · 1 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites xRAG as a baseline.

compressing all content into a single token representation ... the compact latent representation discards explicit links to source documents, making citation grounding difficult and the pipeline opaque to developers.
— Incorporating Q&A Nuggets into Retrieval-Augmented Generation
“As the original xRAG model uses one token to encode an entire passage, which can lead to hallucinations”
— SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression
“Nevertheless, such soft approaches often lack interpretability and transferability, and degrade task performance.”
— CORE-RAG: Lossless Compression for Retrieval-Augmented LLMs via Reinforcement Learning

Beaten on benchmarks

Head-to-head results where a newer method reports beating xRAG. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.