How to Synthesize Text Data without Model Collapse?

TL;DR

This paper investigates the impact of synthetic data on language model training, identifies issues like distributional shift and over-concentration, and proposes token editing of human data as a method to prevent model collapse and improve performance.

Contribution

It introduces a novel token editing technique on human data to synthesize semi-synthetic data, theoretically preventing model collapse during training.

Findings

Higher synthetic data proportions degrade model performance.

Token editing constrains test error with a finite upper bound.

Token editing improves performance across various training scenarios.

Abstract

Model collapse in synthetic data indicates that iterative training on self-generated data leads to a gradual decline in performance. With the proliferation of AI models, synthetic data will fundamentally reshape the web data ecosystem. Future GPT-$\{n\}$ models will inevitably be trained on a blend of synthetic and human-produced data. In this paper, we focus on two questions: what is the impact of synthetic data on language model training, and how to synthesize data without model collapse? We first pre-train language models across different proportions of synthetic data, revealing a negative correlation between the proportion of synthetic data and model performance. We further conduct statistical analysis on synthetic data to uncover distributional shift phenomenon and over-concentration of n-gram features. Inspired by the above findings, we propose token editing on human-produced data to obtain semi-synthetic data. As a proof of concept, we theoretically demonstrate that token-level editing can prevent model collapse, as the test error is constrained by a finite upper bound. We conduct extensive experiments on pre-training from scratch, continual pre-training, and supervised fine-tuning. The results validate our theoretical proof that token-level editing improves model performance.