Code Summarization Using Mamba-SoTaNa

Effectively understanding and managing source code has become increasingly important as modern software systems grow in complexity. This study aims to explore the potential of a model that can maximize memory efficiency in resource-constrained environments and enhance performance on large codebases, providing practical benefits for software development and maintenance. To achieve this, the Mamba model was integrated into SoTaNa, and experiments were conducted to measure code summarization performance using BLEU, ROUGE-L, METEOR, and BERTScore metrics. The performance of the Mamba-SoTaNa model was compared with that of the existing SoTaNa and LLaMA models. The results showed that the Mamba-SoTaNa model outperformed the LLaMA model but did not surpass the performance of the SoTaNa model. In conclusion, this study highlights the need for further research and optimization to improve the code summarization performance of the Mamba model.