---
library_name: transformers
datasets:
- NamCyan/tesoro-code
base_model:
- codellama/CodeLlama-7b-hf
---

# Improving the detection of technical debt in Java source code with an enriched dataset


## Model Details

### Model Description

This model is the part of Tesoro project, used for detecting technical debt in source code. More information can be found at [Tesoro HomePage](https://github.com/NamCyan/tesoro.git).

- **Developed by:** [Nam Hai Le](https://github.com/NamCyan)
- **Model type:** Decoder-based PLMs
- **Language(s):** Java
- **Finetuned from model:** [Codellama](https://huggingface.co./codellama/CodeLlama-7b-hf)

### Model Sources

- **Repository:** [Tesoro](https://github.com/NamCyan/tesoro.git)
- **Paper:** [To be update]

## How to Get Started with the Model

Use the code below to get started with the model.

```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("NamCyan/CodeLlama-7b-technical-debt-code-tesoro")
model = AutoModelForSequenceClassification.from_pretrained("NamCyan/CodeLlama-7b-technical-debt-code-tesoro")
```


## Training Details

- Training Data: The model is finetuned using [tesoro-code](https://huggingface.co./datasets/NamCyan/tesoro-code)

- Infrastructure: Training process is conducted on two NVIDIA A100 GPUs with 80GB of VRAM. [LoRa](https://arxiv.org/abs/2106.09685) is adopted to train this model.


## Leaderboard
| Model | Model size  | EM            | F1              |
|:-------------|:-----------|:------------------|:------------------|
| **Encoder-based PLMs** |
| [CodeBERT](https://huggingface.co./microsoft/codebert-base)     | 125M       | 38.28             | 43.47             |
| [UniXCoder](https://huggingface.co./microsoft/unixcoder-base)    | 125M       | 38.12             | 42.58             |
| [GraphCodeBERT](https://huggingface.co./microsoft/graphcodebert-base)| 125M       | *39.38*          | *44.21*           |
| [RoBERTa](https://huggingface.co./FacebookAI/roberta-base)      | 125M       | 35.37             | 38.22             |
| [ALBERT](https://huggingface.co./albert/albert-base-v2)       | 11.8M      | 39.32             | 41.99             |
| **Encoder-Decoder-based PLMs** |
| [PLBART](https://huggingface.co./uclanlp/plbart-base)       | 140M       | 36.85             | 39.90             |
| [Codet5](https://huggingface.co./Salesforce/codet5-base)       | 220M       | 32.66             | 35.41             |
| [CodeT5+](https://huggingface.co./Salesforce/codet5p-220m)      | 220M       | 37.91             | 41.96             |
| **Decoder-based PLMs (LLMs)** |
| [TinyLlama](https://huggingface.co./TinyLlama/TinyLlama_v1.1_math_code)    | 1.03B      | 37.05             | 40.05             |
| [DeepSeek-Coder](https://huggingface.co./deepseek-ai/deepseek-coder-1.3b-base) | 1.28B    | **42.52**         | **46.19**         |
| [OpenCodeInterpreter](https://huggingface.co./m-a-p/OpenCodeInterpreter-DS-1.3B)       | 1.35B             | 38.16             | 41.76             |
| [phi-2](https://huggingface.co./microsoft/phi-2)        | 2.78B      | 37.92             | 41.57             |
| [starcoder2](https://huggingface.co./bigcode/starcoder2-3b)   | 3.03B      | 35.37             | 41.77             |
| [CodeLlama](https://huggingface.co./codellama/CodeLlama-7b-hf)    | 6.74B      | 34.14             | 38.16             |
| [Magicoder](https://huggingface.co./ise-uiuc/Magicoder-S-DS-6.7B)    | 6.74B      | 39.14             | 42.49             |


## Citing us
```bibtex
@article{nam2024tesoro,
  title={Improving the detection of technical debt in Java source code with an enriched dataset},
  author={Hai, Nam Le and Bui, Anh M. T. Bui and Nguyen, Phuong T. and Ruscio, Davide Di and Kazman, Rick},
  journal={},
  year={2024}
}
```