metadata
license: apache-2.0
datasets:
- rgeada/tool_response_injections
metrics:
- f1
- accuracy
base_model:
- answerdotai/ModernBERT-base
tags:
- Guardrail
- Tool
- Classifier
Tool Response Injection Detector
ModernBert fine-tuned over rgeada/tool_response_injections. Intended for use with NeMo Guardrails' HuggingFace Classifier rails, as a tool-response injection guardrail.
Scores:
- Test F1-Score: 0.8464
- Test Accuracy: 0.8466