# PLBART

**Repository Path**: ecust-dp/plbart

## Basic Information

- **Project Name**: PLBART
- **Description**: Code replication for Vulnerability Identification experiments in the NAACL 2021 paper: Unified Pre-training for Program Understanding and Generation
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-06-12
- **Last Updated**: 2024-06-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

#### 项目介绍
Code replication for Vulnerability Identification experiments in the NAACL 2021 paper: Unified Pre-training for Program Understanding and Generation

#### 系统环境

**P40-1**_高性能计算平台应用名称：gym

    CPU：10 核
    
    RAM：100 GB
    
    GPU：NVIDIA Tesla P40 24G
    
    OS：Ubuntu 18.04


**P40-2**_高性能计算平台应用名称：Tensorflow_GPU

    CPU：10 核
    
    RAM：64 GB 
    
    GPU：NVIDIA Tesla P40 24G
    
    OS：CentOS 7.7


**3090**_高性能计算平台应用名称：Desktop_GPU

    CPU：10 核
    
    RAM：20 GB
    
    GPU：Nvidia GeForce RTX 3090
    
    OS：CentOS 7.8


#### 环境配置

```
git clone https://gitee.com/ecust-dp/plbart.git
cd plbart
conda create --name PLBART python=3.6.10
conda activate PLBART
conda install pytorch==1.5.1 torchvision==0.6.1 cudatoolkit=10.2
pip install submitit
```

```
mkdir third_party
cd third_party
```

```git clone https://gitee.com/ecust-dp/fairseq``` or ```git clone https://github.com/pytorch/fairseq```

```
cd fairseq
git checkout 698e3b91ffa832c286c48035bdff78238b0de8ae
pip install .
cd ../..
```

```
pip install sacrebleu==1.2.11
pip install tree_sitter==0.2.1
pip install sentencepiece
pip install scikit-learn
```

**Note:** refer to **requirements_notes.txt** if encountered error during installing packages mentioned above.


**Prepare pretrained PLBART**

Download **plbart_base.pt** from 高性能计算平台(Path: Ecust-SE/DP/PLBART-JIT/pretrain/), and put it under the **pretrain/** folder.

**Prepare data files for fine-tuning PLBART**

Download **code-to-code.zip** from 高性能计算平台(Path: Ecust-SE/DP/PLBART-JIT/data/codeXglue), and upload it to **data/codeXglue**. Extract the **code-to-code** folder containing **defect_prediction** folder via ```unzip code-to-code.zip```

#### 使用说明

**preprocess data and run**
```
cd scripts/code_to_code/defect_prediction/

bash prepare.sh

bash run.sh 0
```


#### 结果对比

Table 9: Results on the vulnerable code detection (accuracy) task.

| Methods      | Acc-paper | Acc-4090 | Acc-3090 | Acc-P40-1 | Acc-P40-2    |
|--------------|-----------|----------|----------|-----------|--------------|
| Transformer  | 61.64     |          | 54.06    | 54.06     | 54.06        |
| CodeBERT     | 62.08     |          | 63.91    | 63.54     | 63.54        |
| PLBART       | 63.18     | 61.20    | 61.75    | 61.79     | 61.79        |

**Notes:** 
1. Other metrics such as precision, recall, and fl-score can be found in **results_GPU_Type.png**
2. Detailed log regrading environment installing and fine-tuning PLBART is recorded in **prepare_environments_running_P40-2.log**.