GPT-4o as the Gold Standard: A Scalable and General Purpose Approach to Filter Language Model Pretraining Data

2024-10-03

AI Summary

To address the high cost and low efficiency of curating high-quality web data for large language model training, this paper proposes SIEVE: a novel active learning–knowledge distillation framework guided by GPT-4o to train lightweight text classifiers. SIEVE achieves low-cost (<1% API call overhead) and high-accuracy data filtering while supporting few-shot annotation, multi-prompt strategies, and domain adaptation—enhancing generalizability and scalability. Evaluated on five specialized filtering tasks, it matches GPT-4o’s performance; in the DataComp-LM benchmark, it surpasses prior state-of-the-art methods and yields substantive improvements in downstream model quality. Its core contribution lies in the first principled balance between large-model capability and lightweight-model efficiency, establishing an efficient, general-purpose, and customizable paradigm for high-fidelity pretraining data curation.

Technology Category

• Machine Learning: Active Learning
• Data Mining & Knowledge Management: Knowledge Acquisition from the Web
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Large language models require vast amounts of high-quality training data, but effective filtering of web-scale datasets remains a significant challenge. This paper demonstrates that GPT-4o is remarkably effective at identifying high-quality training data, but its prohibitive cost makes it impractical at web-scale. We propose SIEVE, a lightweight alternative that matches GPT-4o accuracy at less than 1% of the cost. SIEVE can perform up to 500 filtering operations for the cost of one GPT-4o filtering call. The key to SIEVE is a seamless integration of GPT-4o and lightweight text classification models, using active learning to fine-tune these models in the background with a small number of calls to GPT-4o. Once trained, it performs as well as GPT-4o at a tiny fraction of the cost. Through different filtering prompts, SIEVE can efficiently curate high quality data for general or specialized domains from web-scale corpora -- a valuable capability given the current scarcity of high-quality domain-specific datasets. Extensive experiments using automatic and human evaluation metrics show that SIEVE and GPT-4o achieve similar performance on five highly specific filtering prompts. In addition, when performing quality filtering on web crawl datasets, we demonstrate SIEVE can further improve over state-of-the-art quality filtering methods in the DataComp-LM challenge for selecting LLM pretraining data.

Problem

• Large-scale Language Models
• Data Selection
• Internet Data Quality

Innovation

• SIEVE
• active learning
• high-quality training data