Skip to content

ProtectAI

any_guardrail.guardrails.protectai.protectai

Protectai

Bases: HuggingFace

Prompt injection detection encoder based models.

For more information, please see the model card:

Source code in src/any_guardrail/guardrails/protectai/protectai.py 
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
class Protectai(HuggingFace):
    """Prompt injection detection encoder based models.

    For more information, please see the model card:

    - [ProtectAI](https://huggingface.co/collections/protectai/llm-security-65c1f17a11c4251eeab53f40).
    """

    SUPPORTED_MODELS: ClassVar = [
        "ProtectAI/deberta-v3-small-prompt-injection-v2",
        "ProtectAI/distilroberta-base-rejection-v1",
        "ProtectAI/deberta-v3-base-prompt-injection",
        "ProtectAI/deberta-v3-base-prompt-injection-v2",
    ]

    def _post_processing(self, model_outputs: dict[str, Any]) -> GuardrailOutput:
        return _match_injection_label(model_outputs, PROTECTAI_INJECTION_LABEL, self.model.config.id2label)