Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ziarko, Alicja, Jiang, Albert Q., Piotrowski, Bartosz, Li, Wenda, Jamnik, Mateja, Miłoś, Piotr
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913582226079744
author Ziarko, Alicja
Jiang, Albert Q.
Piotrowski, Bartosz
Li, Wenda
Jamnik, Mateja
Miłoś, Piotr
author_facet Ziarko, Alicja
Jiang, Albert Q.
Piotrowski, Bartosz
Li, Wenda
Jamnik, Mateja
Miłoś, Piotr
contents Text embeddings are essential for many tasks, such as document retrieval, clustering, and semantic similarity assessment. In this paper, we study how to contrastively train text embedding models in a compute-optimal fashion, given a suite of pre-trained decoder-only language models. Our innovation is an algorithm that produces optimal configurations of model sizes, data quantities, and fine-tuning methods for text-embedding models at different computational budget levels. The resulting recipe, which we obtain through extensive experiments, can be used by practitioners to make informed design choices for their embedding models. Specifically, our findings suggest that full fine-tuning and low-rank adaptation fine-tuning produce optimal models at lower and higher computational budgets respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2406_04165
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe
Ziarko, Alicja
Jiang, Albert Q.
Piotrowski, Bartosz
Li, Wenda
Jamnik, Mateja
Miłoś, Piotr
Machine Learning
Text embeddings are essential for many tasks, such as document retrieval, clustering, and semantic similarity assessment. In this paper, we study how to contrastively train text embedding models in a compute-optimal fashion, given a suite of pre-trained decoder-only language models. Our innovation is an algorithm that produces optimal configurations of model sizes, data quantities, and fine-tuning methods for text-embedding models at different computational budget levels. The resulting recipe, which we obtain through extensive experiments, can be used by practitioners to make informed design choices for their embedding models. Specifically, our findings suggest that full fine-tuning and low-rank adaptation fine-tuning produce optimal models at lower and higher computational budgets respectively.
title Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe
topic Machine Learning
url https://arxiv.org/abs/2406.04165