Skip to content

P-H-B-D/EmbeddingsTranslation

Repository files navigation

VectorBridge: A Robust Technique for Cross-Model Embeddings Translation

Peter Bowman-Davis and Roshan Palakkal

VectorBridge is aimed at creating an interoperability layer between different sentence embeddings models, thereby promoting vendor neutrality and enabling users to swap models freely based on their specific needs and preferences, without the need for re-embedding a sample corpus. VectorBridge thereby enables the querying vector databases/stores encoded in closed-source models, using only open-source tools.

Overview

/models Contains pretrained .h5 models that can be loaded to translate embeddings.

/benchmarking Contains benchmarks relating to translational models using the MTEB evaluation metrics.

/datasets Is an incomplete repository of data we used to train the models. Any datasets not present in this folder (by nature of their size) will be mentioned in the Methodology section.

Methodology

VectorBridge is a proof-of-concept demonstrating how a small sample of input:output pairs of embeddings models can be used to deduce an appropriate mapping between differing embeddings models, allowing for the "translation" of vector spaces.

VectorBridge is aimed at creating an interoperability layer between different sentence embeddings models, thereby promoting vendor neutrality and enabling users to swap models freely based on their specific needs and preferences.

## Code Example 

Discussion

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

2 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors