To preserve the integrity of the double-blind review process, we have only included the core components of our implementation in the submission and removed any potentially identifying information from the code. Specifically, the training pipeline is implemented in u_train.py and trainer.py. The base class LoRA.py and our proposed methods, RaLoRA and RaLoRA-Pro, are implemented in RaLoRA.py. Additionally, we provide a set of example bash scripts under the scripts/ directory to facilitate reproduction of the experiments reported in the paper. After the review process, we will publicly release the full codebase, including all supplementary materials, on a GitHub repository linked from the final version of the paper.