Training, validation, and held-out test sets for RAG-PIF, a 3-layer prompt
injection firewall for RAG systems. Full pipeline code:
https://github.com/shivateja1253/rag-pif
Trained model:
https://huggingface.co/shivateja1253/rag-pif-roberta
train_augmented.csv
1179
Training split + hard negatives + OOD-style examples (used to train the production Layer 3 checkpoint)