Resume Content Generation Using Llama 2 with Adapters

Publication Date

1-1-2024

Document Type

Conference Proceeding

Publication Title

Proceedings - IEEE 10th International Conference on Big Data Computing Service and Applications, BigDataService 2024

DOI

10.1109/BigDataService62917.2024.00010

First Page

19

Last Page

26

Abstract

This paper presents a novel approach to enhancing the Llama language model for generating customized resumes tailored to domain-specific job descriptions. Unlike traditional methods that rely heavily on extensive fine-tuning, we implement scalable adapter modules to minimize parameter adjustments. This approach preserves the model's inherent ability to generate coherent and contextually appropriate language across diverse tasks, ensuring that its general linguistic capabilities remain intact. Additionally, we employ a prompting strategy to dynamically create a diverse and comprehensive dataset, ensuring high relevance to various job roles. Using a dataset of 10,000 job descriptions and resumes, our approach resulted in a 20% improvement in BLEU scores and a 70% reduction in perplexity compared to the base model. This combination allows for an efficient and effective fine-tuning process, resulting in superior performance in generating job-specific resume content as evidenced by improved BLEU scores and perplexity metrics. Our approach provides a practical solution for scalable production deployment, while maintaining the model's adaptability and robustness across different job domains.

Keywords

adapter, generative AI, Llama, resume

Department

Computer Science

Link to Full Text

Share

COinS