README.md

---
library_name: transformers
license: mit
base_model: microsoft/mdeberta-v3-base
tags:
- prompt-injection
- injection
- security
- llm-security
- generated_from_trainer
metrics:
- accuracy
- precision
- recall
- f1
model-index:
- name: mdeberta-v3-base-prompt-injection
  results: []
---

# mdeberta-v3-base-prompt-injection

This model is a fine-tuned version of [microsoft/mdeberta-v3-base](https://huggingface.co/microsoft/mdeberta-v3-base) on a combination of [jackhhao/jailbreak-classification](https://huggingface.co/datasets/jackhhao/jailbreak-classification), [deepset/prompt-injections](https://huggingface.co/datasets/deepset/prompt-injections/viewer/default/test?views%5B%5D=test), a custom datasets containing known attacks, and injections nested in legitimate content like websites and articles.

## Usage

```Python
from transformers import pipeline

classifier = pipeline(
  "text-classification",
  model="proventra/mdeberta-v3-base-prompt-injection"
)

print(classifier("Your text to scan"))
```

## Use in Proventra Core
[proventra-core](https://github.com/proventra/proventra-core) python library

check out [Proventra](https://www.proventra-ai.com)