CorrSynth -- A Correlated Sampling Method for Diverse Dataset Generation from LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kowshik, Suhas S, Divekar, Abhishek, Malik, Vijit
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915202704867328
author Kowshik, Suhas S
Divekar, Abhishek
Malik, Vijit
author_facet Kowshik, Suhas S
Divekar, Abhishek
Malik, Vijit
contents Large language models (LLMs) have demonstrated remarkable performance in diverse tasks using zero-shot and few-shot prompting. Even though their capabilities of data synthesis have been studied well in recent years, the generated data suffers from a lack of diversity, less adherence to the prompt, and potential biases that creep into the data from the generator model. In this work, we tackle the challenge of generating datasets with high diversity, upon which a student model is trained for downstream tasks. Taking the route of decoding-time guidance-based approaches, we propose CorrSynth, which generates data that is more diverse and faithful to the input prompt using a correlated sampling strategy. Further, our method overcomes the complexity drawbacks of some other guidance-based techniques like classifier-based guidance. With extensive experiments, we show the effectiveness of our approach and substantiate our claims. In particular, we perform intrinsic evaluation to show the improvements in diversity. Our experiments show that CorrSynth improves both student metrics and intrinsic metrics upon competitive baselines across four datasets, showing the innate advantage of our method.
format Preprint
id arxiv_https___arxiv_org_abs_2411_08553
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle CorrSynth -- A Correlated Sampling Method for Diverse Dataset Generation from LLMs
Kowshik, Suhas S
Divekar, Abhishek
Malik, Vijit
Computation and Language
Artificial Intelligence
Machine Learning
Large language models (LLMs) have demonstrated remarkable performance in diverse tasks using zero-shot and few-shot prompting. Even though their capabilities of data synthesis have been studied well in recent years, the generated data suffers from a lack of diversity, less adherence to the prompt, and potential biases that creep into the data from the generator model. In this work, we tackle the challenge of generating datasets with high diversity, upon which a student model is trained for downstream tasks. Taking the route of decoding-time guidance-based approaches, we propose CorrSynth, which generates data that is more diverse and faithful to the input prompt using a correlated sampling strategy. Further, our method overcomes the complexity drawbacks of some other guidance-based techniques like classifier-based guidance. With extensive experiments, we show the effectiveness of our approach and substantiate our claims. In particular, we perform intrinsic evaluation to show the improvements in diversity. Our experiments show that CorrSynth improves both student metrics and intrinsic metrics upon competitive baselines across four datasets, showing the innate advantage of our method.
title CorrSynth -- A Correlated Sampling Method for Diverse Dataset Generation from LLMs
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2411.08553