How to Synthesize Text Data without Model Collapse?

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 blen...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2024-12
Main Authors: Zhu, Xuekai, Cheng, Daixuan, Li, Hengli, Zhang, Kaiyan, Ermo Hua, Lv, Xingtai, Ding, Ning, Lin, Zhouhan, Zheng, Zilong, Bowen, Zhou
Format: Article
Language:English
Subjects:
Editing
Statistical analysis
Synthetic data
Upper bounds
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary: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 data quality and enhances model performance.
ISSN:2331-8422