app.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
import torch


class Guardrail:
    def __init__(self):
        tokenizer = AutoTokenizer.from_pretrained("ProtectAI/deberta-v3-base-prompt-injection")
        model = AutoModelForSequenceClassification.from_pretrained("ProtectAI/deberta-v3-base-prompt-injection")

        self.classifier = pipeline(
            "text-classification",
            model=model,
            tokenizer=tokenizer,
            truncation=True,
            max_length=512,
            device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
        )

    def guard(self, prompt):
        return self.classifier(prompt)


class TextPrompt(BaseModel):
    prompt: str


app = FastAPI()
guardrail = Guardrail()


@app.post("/classify/")
def classify_text(text_prompt: TextPrompt):
    try:
        result = guardrail.guard(text_prompt.prompt)
        return result
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=8000)