---
license: mit
language:
- en
metrics:
- f1
- recall
base_model:
- microsoft/deberta-base
---
## About this Project

- This project aims to build a robust filter that detects malicious prompt injection attacks before it reaches the LLM.
- **The detector can detect a prompt injection attack in few-shot manner, with a fine-tuned Siamese network based on a pretrained DeBERTa model.**
- The detector can prevent prompt injection in the scenario where we have only a few examples of a novel injection attack. The experiment showed that this approach outperformed the existing cross-entropy-based fine-tuning.

Please refer to [GitHub](https://github.com/Superfish83/PromptKiller/) for details.


## Training Info

**Backbone model**: [DeBERTa 86M](https://huggingface.co/microsoft/deberta-base)

**Datasets**: This project used all or part of each dataset shown below.

- https://www.kaggle.com/datasets/marycamilainfo/prompt-injection-malignant
- https://www.kaggle.com/datasets/arielzilber/prompt-injection-in-the-wild
- https://www.kaggle.com/datasets/mohammedaminejebbar/malicious-prompt-detection-dataset-mpdd
- https://www.kaggle.com/datasets/arielzilber/prompt-injection-suffix-attack