# Blaze (Finetuned Variant)

This model is a finetuned version of the **BLAZE** model described in the paper:

**_BLAZE: Cross‑Language and Cross‑Project Bug Localization via Dynamic Chunking and Hard Example Learning_**  
DOI: [10.5281/zenodo.15122980](https://doi.org/10.5281/zenodo.15122980) :contentReference[oaicite:1]{index=1}

---

## 📘 What’s Inside

A Transformer-based bug localization model fine-tuned on additional cross-project datasets to enhance its ability to pinpoint bugs in unseen codebases.

---

## 🧪 Fine-tuning Details

- **Starting point:** Pretrained Blaze model from Zenodo.
- **Enhancements:** Further trained on a curated dataset using dynamic chunking and hard-negative sampling — all detailed in the original manuscript and accompanying code release.

---

## 🔍 Intended Usage

- **Primary task:** Automatic identification of buggy code segments across languages and projects.
- **How to use:** Load the model and feed it code snippets to receive localized bug predictions.

---

## 📄 Citation

If you use this model, please cite:

> _BLAZE: Cross‑Language and Cross‑Project Bug Localization via Dynamic Chunking and Hard Example Learning_, available at DOI: 10.5281/zenodo.15122980 :contentReference[oaicite:2]{index=2}

---

## 📁 Contents of This Repo

- `config.json`  
- `pytorch_model.bin` (or `tf_model.h5`)  
- `tokenizer.json` (if applicable)  
- `README.md`

---

## 🛠️ Setup

```bash
pip install transformers huggingface_hub