transformers - 💡(How to fix) Fix New RAG framework: VORTEXRAG — 7-layer causal RAG achieving EM 74.8, Faithfulness 0.94

Hi Transformers team!

Sharing VORTEXRAG — a new RAG framework built on top of HuggingFace Transformers.

VORTEXRAG uses:

• all-mpnet-base-v2 (sentence-transformers) for the 768d semantic arm
• DeBERTa-v3-small CrossEncoder for the Faithfulness Verifier (FV) layer
• spaCy + PropBank for syntactic/causal arms

The framework adds 7 layers of causal filtering on top of these models to eliminate semantic drift and context window poisoning.

Results: EM 74.8, F1 82.6, Faithfulness 0.94 across NQ/TriviaQA/WebQ/PopQA/HotpotQA/2WikiMH

Would love to know if there's a good way to contribute this to the HuggingFace ecosystem.

• GitHub: https://github.com/vignesh2027/VORTEXRAG
• Paper: https://doi.org/10.5281/zenodo.20579702
• HF Space: https://huggingface.co/spaces/vigneshwar234/VORTEXRAG