How Can We Synthesize High-Quality Pretraining Data? A Systematic Study of Prompt Design, Generator Model, and Source Data

• URL: http://arxiv.org/abs/2604.13977v1
• Date: Wed, 15 Apr 2026 15:24:59 GMT
• Title: How Can We Synthesize High-Quality Pretraining Data? A Systematic Study of Prompt Design, Generator Model, and Source Data
• Abstract summary: We identify critical factors in rephrasing web text into synthetic pretraining data.<n> structured output formats consistently outperform both curated web baselines and prior synthetic methods.<n>We develop textbftextscFinePhrase, a 486-billion-token open dataset of rephrased web text.
• Score: 27.516126966056632
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Synthetic data is a standard component in training large language models, yet systematic comparisons across design dimensions, including rephrasing strategy, generator model, and source data, remain absent. We conduct extensive controlled experiments, generating over one trillion tokens, to identify critical factors in rephrasing web text into synthetic pretraining data. Our results reveal that structured output formats, such as tables, math problems, FAQs, and tutorials, consistently outperform both curated web baselines and prior synthetic methods. Notably, increasing the size of the generator model beyond 1B parameters provides no additional benefit. Our analysis also demonstrates that the selection of the original data used for mixing substantially influences performance. By applying our findings, we develop \textbf{\textsc{FinePhrase}}, a 486-billion-token open dataset of rephrased web text. We show that \textsc{FinePhrase} outperforms all existing synthetic data baselines while reducing generation costs by up to 30 times. We provide the dataset, all prompts, and the generation framework to the research community.

Related papers

• Demystifying Synthetic Data in LLM Pre-training: A Systematic Study of Scaling Laws, Benefits, and Pitfalls [25.294408301653576]
  Training data plays a crucial role in Large Language Models (LLM) scaling, yet high quality data is of limited supply.<n>We compare natural web data, diverse synthetic types (rephrased text, generated textbooks), and mixtures of natural and synthetic data.<n>We find pre-training on rephrased synthetic data textitalone is not faster than pre-training on natural web texts.
  arXiv  Detail & Related papers  (2025-10-02T03:24:42Z)
• Understanding the Influence of Synthetic Data for Text Embedders [52.04771455432998]
  We first reproduce and publicly release the synthetic data proposed by Wang et al.<n>We critically examine where exactly synthetic data improves model generalization.<n>Our findings highlight the limitations of current synthetic data approaches for building general-purpose embedders.
  arXiv  Detail & Related papers  (2025-09-07T19:28:52Z)
• Synthline: A Product Line Approach for Synthetic Requirements Engineering Data Generation using Large Language Models [0.5156484100374059]
  This paper introduces Synthline, a Product Line (PL) approach that leverages Large Language Models to generate synthetic Requirements Engineering (RE) data.<n>Our analysis reveals that while synthetic datasets exhibit less diversity than real data, they are good enough to serve as viable training resources.<n>Our evaluation shows that combining synthetic and real data leads to substantial performance improvements.
  arXiv  Detail & Related papers  (2025-05-06T07:57:16Z)
• Assessing Generative Models for Structured Data [0.0]
  This paper introduces rigorous methods for assessing synthetic data against real data by looking at inter-column dependencies within the data.<n>We find that large language models (GPT-2), both when queried via few-shot prompting, and when fine-tuned, and GAN (CTGAN) models do not produce data with dependencies that mirror the original real data.
  arXiv  Detail & Related papers  (2025-03-26T18:19:05Z)
• Scaling Laws of Synthetic Data for Language Models [125.41600201811417]
  We introduce SynthLLM, a scalable framework that transforms pre-training corpora into diverse, high-quality synthetic datasets.<n>Our approach achieves this by automatically extracting and recombining high-level concepts across multiple documents using a graph algorithm.
  arXiv  Detail & Related papers  (2025-03-25T11:07:12Z)
• DreamMask: Boosting Open-vocabulary Panoptic Segmentation with Synthetic Data [61.62554324594797]
  We propose DreamMask, which explores how to generate training data in the open-vocabulary setting, and how to train the model with both real and synthetic data.<n>In general, DreamMask significantly simplifies the collection of large-scale training data, serving as a plug-and-play enhancement for existing methods.<n>For instance, when trained on COCO and tested on ADE20K, the model equipped with DreamMask outperforms the previous state-of-the-art by a substantial margin of 2.1% mIoU.
  arXiv  Detail & Related papers  (2025-01-03T19:00:00Z)
• GRIP: A Graph-Based Reasoning Instruction Producer [47.80560026838563]
  We present textittextbfGRIP, a textbfGraph-based textbfReasoning textbfInstruction textbfProducer that efficiently synthesizes high-quality and diverse reasoning instructions.
  arXiv  Detail & Related papers  (2024-12-12T01:52:25Z)
• Exploring the Landscape for Generative Sequence Models for Specialized Data Synthesis [0.0]
  This paper introduces a novel approach that leverages three generative models of varying complexity to synthesize Malicious Network Traffic. Our approach transforms numerical data into text, re-framing data generation as a language modeling task. Our method surpasses state-of-the-art generative models in producing high-fidelity synthetic data.
  arXiv  Detail & Related papers  (2024-11-04T09:51:10Z)
• Generating Realistic Tabular Data with Large Language Models [49.03536886067729]
  Large language models (LLM) have been used for diverse tasks, but do not capture the correct correlation between the features and the target variable. We propose a LLM-based method with three important improvements to correctly capture the ground-truth feature-class correlation in the real data. Our experiments show that our method significantly outperforms 10 SOTA baselines on 20 datasets in downstream tasks.
  arXiv  Detail & Related papers  (2024-10-29T04:14:32Z)
• Reimagining Synthetic Tabular Data Generation through Data-Centric AI: A Comprehensive Benchmark [56.8042116967334]
  Synthetic data serves as an alternative in training machine learning models. ensuring that synthetic data mirrors the complex nuances of real-world data is a challenging task. This paper explores the potential of integrating data-centric AI techniques to guide the synthetic data generation process.
  arXiv  Detail & Related papers  (2023-10-25T20:32:02Z)
• FairGen: Fair Synthetic Data Generation [0.3149883354098941]
  We propose a pipeline to generate fairer synthetic data independent of the GAN architecture. We claim that while generating synthetic data most GANs amplify bias present in the training data but by removing these bias inducing samples, GANs essentially focuses more on real informative samples.
  arXiv  Detail & Related papers  (2022-10-24T08:13:47Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.