Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dial-insight: Fine-tuning Large Language Models with High-Quality Domain-Specific Data Preventing Capability Collapse

Jianwei Sun, Chaoyang Mei, Linlin Wei, Kaiyu Zheng, Na Liu, Ming Cui, Tianyi Li

TL;DR

It is indicated that the domain-specific proficiency of general LLMs can be enhanced through fine-tuning with data produced via the proposed method, without compromising their overall generalization abilities, even when exclusively domain-specific data is employed for fine-tuning.

Abstract

The efficacy of large language models (LLMs) is heavily dependent on the quality of the underlying data, particularly within specialized domains. A common challenge when fine-tuning LLMs for domain-specific applications is the potential degradation of the model's generalization capabilities. To address these issues, we propose a two-stage approach for the construction of production prompts designed to yield high-quality data. This method involves the generation of a diverse array of prompts that encompass a broad spectrum of tasks and exhibit a rich variety of expressions. Furthermore, we introduce a cost-effective, multi-dimensional quality assessment framework to ensure the integrity of the generated labeling data. Utilizing a dataset comprised of service provider and customer interactions from the real estate sector, we demonstrate a positive correlation between data quality and model performance. Notably, our findings indicate that the domain-specific proficiency of general LLMs can be enhanced through fine-tuning with data produced via our proposed method, without compromising their overall generalization abilities, even when exclusively domain-specific data is employed for fine-tuning.

Dial-insight: Fine-tuning Large Language Models with High-Quality Domain-Specific Data Preventing Capability Collapse

TL;DR

It is indicated that the domain-specific proficiency of general LLMs can be enhanced through fine-tuning with data produced via the proposed method, without compromising their overall generalization abilities, even when exclusively domain-specific data is employed for fine-tuning.

Abstract

The efficacy of large language models (LLMs) is heavily dependent on the quality of the underlying data, particularly within specialized domains. A common challenge when fine-tuning LLMs for domain-specific applications is the potential degradation of the model's generalization capabilities. To address these issues, we propose a two-stage approach for the construction of production prompts designed to yield high-quality data. This method involves the generation of a diverse array of prompts that encompass a broad spectrum of tasks and exhibit a rich variety of expressions. Furthermore, we introduce a cost-effective, multi-dimensional quality assessment framework to ensure the integrity of the generated labeling data. Utilizing a dataset comprised of service provider and customer interactions from the real estate sector, we demonstrate a positive correlation between data quality and model performance. Notably, our findings indicate that the domain-specific proficiency of general LLMs can be enhanced through fine-tuning with data produced via our proposed method, without compromising their overall generalization abilities, even when exclusively domain-specific data is employed for fine-tuning.
Paper Structure (12 sections, 5 equations, 3 figures, 5 tables)

This paper contains 12 sections, 5 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Two-stage Evolutionary Prompt Data Production Method
  3. Stage 1. Task Instruction Evolution
  4. Stage 2. Prompt Evolution
  5. Data Label Generation
  6. Experiment & Analysis
  7. Data Set Construction
  8. Model training
  9. Data Comprehensive Quality Evaluation
  10. Model performance evaluation
  11. Experimental results
  12. Conclusion

Figures (3)

  • Figure 1: The two-stage Prompt evolution method.
  • Figure 2: Stage 2: Prompt Evolution method
  • Figure 3: The distribution statistics of task types (a) and the distribution statistics of task output formats (b) in the test dataset. Please note that the units on the vertical axis represent the number of test data samples.