Comparative Analysis of GPT-2 and Gemini for Automated Generation of Dockerfiles and Docker Compose in DevOps Workflows

Abstract

With the growing complexity of containerized applications, ensuring efficient and error-free automation has become a critical challenge in modern software deployment workflows. This work compares GPT-2 and Gemini for Dockerfile and Docker Compose generation by examining three use cases of increasing complexity. The results demonstrat that although GPT-2 can produce valid configurations after fine-tuning, it often requires significant manual adjustments to meet specific requirements. On the other hand, Gemini demonstrated its abilty to generate complete and correct configurations with minimal or no human intervention. The quantitative scores obtained with BLEU and ROUGE-L also confirm that Gemini produces better quality results. Overall, the results of this study confirm Gemini's potential to change DevOps workflows, particularly for production environments requiring both reliability and efficiency. These results highlight Gemini's potential as a transformative tool in DevOps workflows, particularly for automating configuration generation in high-stakes environments. The study also highlights opportunities for future research, such as exploring the potential benefits of using Gemini for specialized tasks like Kubernetes orchestration.