Retro-li: Small-Scale Retrieval Augmented Generation Supporting Noisy Similarity Searches and Domain Shift Generalization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rashiti, Gentiana, Karunaratne, Geethan, Sachan, Mrinmaya, Sebastian, Abu, Rahimi, Abbas
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912294271713280
author Rashiti, Gentiana
Karunaratne, Geethan
Sachan, Mrinmaya
Sebastian, Abu
Rahimi, Abbas
author_facet Rashiti, Gentiana
Karunaratne, Geethan
Sachan, Mrinmaya
Sebastian, Abu
Rahimi, Abbas
contents The retrieval augmented generation (RAG) system such as Retro has been shown to improve language modeling capabilities and reduce toxicity and hallucinations by retrieving from a database of non-parametric memory containing trillions of entries. We introduce Retro-li that shows retrieval can also help using a small-scale database, but it demands more accurate and better neighbors when searching in a smaller hence sparser non-parametric memory. This can be met by using a proper semantic similarity search. We further propose adding a regularization to the non-parametric memory for the first time: it significantly reduces perplexity when the neighbor search operations are noisy during inference, and it improves generalization when a domain shift occurs. We also show that Retro-li's non-parametric memory can potentially be implemented on analog in-memory computing hardware, exhibiting O(1) search time while causing noise in retrieving neighbors, with minimal (<1%) performance loss. Our code is available at: https://github.com/IBM/Retrieval-Enhanced-Transformer-Little.
format Preprint
id arxiv_https___arxiv_org_abs_2410_00004
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Retro-li: Small-Scale Retrieval Augmented Generation Supporting Noisy Similarity Searches and Domain Shift Generalization
Rashiti, Gentiana
Karunaratne, Geethan
Sachan, Mrinmaya
Sebastian, Abu
Rahimi, Abbas
Information Retrieval
Artificial Intelligence
Computation and Language
The retrieval augmented generation (RAG) system such as Retro has been shown to improve language modeling capabilities and reduce toxicity and hallucinations by retrieving from a database of non-parametric memory containing trillions of entries. We introduce Retro-li that shows retrieval can also help using a small-scale database, but it demands more accurate and better neighbors when searching in a smaller hence sparser non-parametric memory. This can be met by using a proper semantic similarity search. We further propose adding a regularization to the non-parametric memory for the first time: it significantly reduces perplexity when the neighbor search operations are noisy during inference, and it improves generalization when a domain shift occurs. We also show that Retro-li's non-parametric memory can potentially be implemented on analog in-memory computing hardware, exhibiting O(1) search time while causing noise in retrieving neighbors, with minimal (<1%) performance loss. Our code is available at: https://github.com/IBM/Retrieval-Enhanced-Transformer-Little.
title Retro-li: Small-Scale Retrieval Augmented Generation Supporting Noisy Similarity Searches and Domain Shift Generalization
topic Information Retrieval
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2410.00004