Supplementary Material

This package contains supplementary material for the paper "Code Comment Inconsistency Detection and Rectification Using a Large Language Model". The package is organized as follows:

• Data.7z: Contains the data used in our study.
• templates/: Contains the LLaMA templates used in our study.
• utils/BalanceTrainer.py: The loss function used in our study.
• utils/prompter.py: The prompter used in our study.
• train.py: The training script used in our study.
• test.py: The testing script used in our study.

run train.py to train the model.

python -u train.py --base_model codellama/CodeLlama-7b-hf \
--data_path Data/LLMtrainDataset.jsonl --output_dir ./LoraCodeLlama_7B --batch_size 32 --micro_batch_size 2 \
--num_epochs 10 --learning_rate 1e-4 --cutoff_len 2048 --val_set_size 100 --prompt_template_name llama \
--label_smoothing_factor 0.1 --classification_alpha 0.5 --train_on_inputs False

run test.py to test the model.

 python -u test.py --base_model codellama/CodeLlama-7b-hf --lora_weights ./LoraCodeLlama_7B --prompt_template llama