This project provides the dataset (SysRetar) and the fine-tuned model (SysRetar-LLM) in Boosting Large Language Models for System Software Retargeting: A Preliminary Study.

Tesyn is a template synthesis approach for prompt construction to enhance LLMs’ performance in system software retargeting.

SysRetar is a dataset specialized for system software retargeting. It consists of four kinds of open-source system software, including two compilers, LLVM and GCC, a hypervisor, xvisor, and a C language library, musl. They can be used to assess the efficacy of SysRetar-LLM across different types of system software and different software (GCC and LLVM) within the same type (compiler).

The composition of SysRetar is provided as follows:

We fine-tuned CodeLLaMA-7b-Instruct to yield SysRetar-LLM.

You can fine-tune CodeLLaMA-7b-Instruct on our datasets by running:

Our fine-tuned SysRetar-LLM is saved in ./Saved_Models/*.

Run following command for inferencing:

The SysRetar-LLM-generated code will be saved in ./Script/Model_Res.

Run following command to calculate the BLEU-4, Edit Distance and CodeBERTScore for generated code:

The results will be saved in ./Script/Result.

@inproceedings{zhong2025tesyn,
  title={Boosting Large Language Models for System Software Retargeting: A Preliminary Study},
  author={Ming Zhong, Fang Lv, Lulin Wang, Lei Qiu, Hongna Geng, Huimin Cui, Xiaobing Feng},
  booktitle={2025 IEEE 32nd International Conference on Software Analysis, Evolution and Reengineering (SANER)},
  year={2025}
}