Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation

Gauthier Guinet, Behrooz Omidvar-Tehrani, Anoop Deoras, Laurent Callot
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:16773-16801, 2024.

Abstract

We propose a new method to measure the task-specific accuracy of Retrieval-Augmented Large Language Models (RAG). Evaluation is performed by scoring the RAG on an automatically-generated synthetic exam composed of multiple choice questions based on the corpus of documents associated with the task. Our method is an automated, cost-efficient, interpretable, and robust strategy to select the optimal components for a RAG system. We leverage Item Response Theory (IRT) to estimate the quality of an exam and its informativeness on task-specific accuracy. IRT also provides a natural way to iteratively improve the exam by eliminating the exam questions that are not sufficiently informative about a model’s ability. We demonstrate our approach on four new open-ended Question-Answering tasks based on Arxiv abstracts, StackExchange questions, AWS DevOps troubleshooting guides, and SEC filings. In addition, our experiments reveal more general insights into factors impacting RAG performance like size, retrieval mechanism, prompting and fine-tuning. Most notably, our findings show that choosing the right retrieval algorithms often leads to bigger performance gains than simply using a larger language model.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-guinet24a, title = {Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation}, author = {Guinet, Gauthier and Omidvar-Tehrani, Behrooz and Deoras, Anoop and Callot, Laurent}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {16773--16801}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/guinet24a/guinet24a.pdf}, url = {https://proceedings.mlr.press/v235/guinet24a.html}, abstract = {We propose a new method to measure the task-specific accuracy of Retrieval-Augmented Large Language Models (RAG). Evaluation is performed by scoring the RAG on an automatically-generated synthetic exam composed of multiple choice questions based on the corpus of documents associated with the task. Our method is an automated, cost-efficient, interpretable, and robust strategy to select the optimal components for a RAG system. We leverage Item Response Theory (IRT) to estimate the quality of an exam and its informativeness on task-specific accuracy. IRT also provides a natural way to iteratively improve the exam by eliminating the exam questions that are not sufficiently informative about a model’s ability. We demonstrate our approach on four new open-ended Question-Answering tasks based on Arxiv abstracts, StackExchange questions, AWS DevOps troubleshooting guides, and SEC filings. In addition, our experiments reveal more general insights into factors impacting RAG performance like size, retrieval mechanism, prompting and fine-tuning. Most notably, our findings show that choosing the right retrieval algorithms often leads to bigger performance gains than simply using a larger language model.} }
Endnote
%0 Conference Paper %T Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation %A Gauthier Guinet %A Behrooz Omidvar-Tehrani %A Anoop Deoras %A Laurent Callot %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-guinet24a %I PMLR %P 16773--16801 %U https://proceedings.mlr.press/v235/guinet24a.html %V 235 %X We propose a new method to measure the task-specific accuracy of Retrieval-Augmented Large Language Models (RAG). Evaluation is performed by scoring the RAG on an automatically-generated synthetic exam composed of multiple choice questions based on the corpus of documents associated with the task. Our method is an automated, cost-efficient, interpretable, and robust strategy to select the optimal components for a RAG system. We leverage Item Response Theory (IRT) to estimate the quality of an exam and its informativeness on task-specific accuracy. IRT also provides a natural way to iteratively improve the exam by eliminating the exam questions that are not sufficiently informative about a model’s ability. We demonstrate our approach on four new open-ended Question-Answering tasks based on Arxiv abstracts, StackExchange questions, AWS DevOps troubleshooting guides, and SEC filings. In addition, our experiments reveal more general insights into factors impacting RAG performance like size, retrieval mechanism, prompting and fine-tuning. Most notably, our findings show that choosing the right retrieval algorithms often leads to bigger performance gains than simply using a larger language model.
APA
Guinet, G., Omidvar-Tehrani, B., Deoras, A. & Callot, L.. (2024). Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:16773-16801 Available from https://proceedings.mlr.press/v235/guinet24a.html.

Related Material