Add initial model card for MVP: Multi-view-guided Passage Reranking

README.md

@@ -0,0 +1,73 @@
+---
+license: apache-2.0
+pipeline_tag: text-ranking
+library_name: transformers
+base_model: google/t5-3b
+---
+
+# MVP: Multi-view-guided Passage Reranking with Large Language Models
+
+This repository contains the official implementation of the paper: [**Multi-view-guided Passage Reranking with Large Language Models**](https://huggingface.co/papers/2509.07485).
+
+<div align="center">
+
+<a href="https://huggingface.co/papers/2509.07485"><img src="https://img.shields.io/badge/arXiv-Paper-red" alt="arXiv Paper"></a>
+<a href="https://github.com/bulbna/MVP"><img src="https://img.shields.io/badge/GitHub-Code-blue?logo=github" alt="GitHub Code"></a>
+
+</div>
+
+## Overview
+<p align="center">
+  <img src="https://github.com/bulbna/MVP/raw/main/assets/fig_MVP_motivation.png" alt="MVP Motivation" width="50%">
+</p>
+
+Recent advances in large language models (LLMs) have shown impressive performance in passage reranking tasks. Despite their success, LLM-based methods still face challenges in efficiency and sensitivity to external biases.
+- (i) Existing models rely mostly on autoregressive generation and sliding window strategies to rank passages, which incurs heavy computational overhead as the number of passages increases.
+- (ii) External biases, such as positional or semantic bias, hinder the model’s ability to accurately represent passages and the input-order sensitivity.
+
+To address these limitations, we propose Multi-View-guided Passage Reranking (MVP), a non-generative LLM-based reranker that encodes query–passage information into multiple views and computes relevance scores via anchor vectors in a single decoding step. An orthogonal loss encourages diversity across views. With only 220M parameters, MVP matches 7B-scale fine-tuned models while reducing inference latency by 100×, and the 3B variant achieves state-of-the-art results on both in-domain and out-of-domain benchmarks.
+
+## Setup Environment
+```
+conda env create -f mvp.yaml
+conda activate mvp
+```
+
+## How to Use
+### Run MVP
+```
+cd inference
+bash run_evaluation.sh
+```
+
+### Train MVP
+```
+cd train
+bash train.sh
+```
+
+## Model Checkpoints
+1. [MVP-base](https://huggingface.co/Jun421/MVP-base)
+2. [MVP-3b](https://huggingface.co/Jun421/MVP-3b)
+
+## Dataset
+### Evaluation Datasets
+- [BM25-Top100](https://huggingface.co/datasets/Soyoung97/beir-eval-bm25-top100)
+
+### Training Datasets
+- [Train/Valid](https://huggingface.co/datasets/Jun421/MVP-train)
+This dataset is derived from BEIR/MSMARCO license, and its usage is restricted to **academic purposes** only.
+
+## Acknowledgments
+We implemented our model based on the following repository: [ListT5](https://github.com/soyoung97/ListT5)
+
+## Citation
+If you find our work useful or helpful for your research, please consider citing our paper:
+```bibtex
+@inproceedings{na2025multiviewguided,
+  title={Multi-view-guided Passage Reranking with Large Language Models},
+  author={Na, Jeongwoo and Kwon, Jun and Choi, Eunseong and Lee, Jongwuk},
+  booktitle={Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)},
+  year={2025}
+}
+```