DiffLM: Controllable Synthetic Data Generation via Diffusion Language Models

2024-11-05

arXiv.org

AI Summary

Large language models (LLMs) suffer from weak modeling of target distributions and cumbersome prompt engineering when synthesizing structured data (e.g., tables, code, tool calls). Method: This paper proposes a controllable synthesis framework based on variational autoencoders (VAEs), introducing diffusion modeling into the language latent space for high-fidelity joint distribution modeling of structured data. It further designs a plug-and-play latent feature injection module that decouples distribution learning from generation control. Contribution/Results: The framework achieves simultaneous preservation of syntactic constraints and semantic distributions without complex prompting. Evaluated on seven real-world structured datasets, it improves downstream task performance by 2–7 percentage points over ground-truth data, demonstrating significant gains in synthetic data quality, controllability, and generalization.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent advancements in large language models (LLMs) have significantly enhanced their knowledge and generative capabilities, leading to a surge of interest in leveraging LLMs for high-quality data synthesis. However, synthetic data generation via prompting LLMs remains challenging due to LLMs' limited understanding of target data distributions and the complexity of prompt engineering, especially for structured formatted data. To address these issues, we introduce DiffLM, a controllable data synthesis framework based on variational autoencoder (VAE), which further (1) leverages diffusion models to reserve more information of original distribution and format structure in the learned latent distribution and (2) decouples the learning of target distribution knowledge from the LLM's generative objectives via a plug-and-play latent feature injection module. As we observed significant discrepancies between the VAE's latent representations and the real data distribution, the latent diffusion module is introduced into our framework to learn a fully expressive latent distribution. Evaluations on seven real-world datasets with structured formatted data (i.e., Tabular, Code and Tool data) demonstrate that DiffLM generates high-quality data, with performance on downstream tasks surpassing that of real data by 2-7 percent in certain cases. The data and code will be publicly available upon completion of internal review.

Problem

• Enhances synthetic data generation via diffusion models
• Improves understanding of target data distributions in LLMs
• Decouples distribution learning from generative objectives

Innovation

• Uses diffusion models for latent distribution learning
• Decouples knowledge learning via plug-and-play module
• Integrates latent diffusion to enhance representation accuracy