---
title: Evidence Graph Reranker
emoji: 🧩
colorFrom: blue
colorTo: indigo
sdk: gradio
app_file: app.py
pinned: false
license: mit
tags:
- rag
- reranking
- multi-hop-qa
- evidence-retrieval
---

# Evidence-Graph Reranker — Hugging Face Space

**A query-conditioned, training-free, joint evidence-subset reranker.** This Space scores query–passage pairs using the public MS MARCO MiniLM-L6 cross-encoder or all-MiniLM-L6 dense encoder, then selects passages with the **same fixed graph algorithm** evaluated on six evidence benchmarks.

For installation, API examples, full six-dataset results and caveats, see [PACKAGE_README.md](PACKAGE_README.md), [MODEL_CARD.md](MODEL_CARD.md), and [results/methodology.md](results/methodology.md).

**Not a new pretrained language model:** This repository provides an algorithm plus an optional Space. It downloads official neural scorer checkpoints on first use; no original model weights are redistributed.