Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Matryoshka-Adaptor: Unsupervised and Supervised Tuning for Smaller Embedding Dimensions

About

Embeddings from Large Language Models (LLMs) have emerged as critical components in various applications, particularly for information retrieval. While high-dimensional embeddings generally demonstrate superior performance as they contain more salient information, their practical application is frequently hindered by elevated computational latency and the associated higher cost. To address these challenges, we propose Matryoshka-Adaptor, a novel tuning framework designed for the customization of LLM embeddings. Matryoshka-Adaptor facilitates substantial dimensionality reduction while maintaining comparable performance levels, thereby achieving a significant enhancement in computational efficiency and cost-effectiveness. Our framework directly modifies the embeddings from pre-trained LLMs which is designed to be seamlessly integrated with any LLM architecture, encompassing those accessible exclusively through black-box APIs. Also, it exhibits efficacy in both unsupervised and supervised learning settings. A rigorous evaluation conducted across a diverse corpus of English, multilingual, and multimodal datasets consistently reveals substantial gains with Matryoshka-Adaptor. Notably, with Google and OpenAI Embedding APIs, Matryoshka-Adaptor achieves a reduction in dimensionality ranging from two- to twelve-fold without compromising performance across multiple BEIR datasets.

Jinsung Yoon, Raj Sinha, Sercan O Arik, Tomas Pfister• 2024

Related benchmarks

TaskDatasetResultRank
Information RetrievalBEIR (test)--
126
Information RetrievalFIQA BEIR (test)
nDCG@1012.97
44
Information RetrievalNQ BEIR
nDCG@1052.44
20
Vision-Language Diagnostic EvaluationCOCO Flickr30K pool
Caption R@138.97
11
Information RetrievalBEIR scidocs (test)
nDCG@100.0387
10
Information RetrievalBEIR scifact (test)
nDCG@1043.43
5
Information RetrievalBEIR arguana (test)
nDCG@1032.68
5
Information RetrievalBEIR quora (test)
nDCG@1077.2
5
Information RetrievalBEIR nfcorpus (test)
nDCG@109.98
5
Showing 9 of 9 rows

Other info

Follow for update