RankMistral100 Open-Source Sorting Model - Free deployment, efficiently re-sort 100 paragraphs with one click

Rankmistral100

Developed by liuwenhan

RankMistral100 is a full-ranking model fine-tuned from Mistral-7B-Instruct-v0.3, specifically designed to re-rank 100 passages in one go, eliminating the need for traditional sliding window strategies.

Large Language Model

Safetensors

EnglishOpen Source License:MIT #Full Ranking of 100 Passages #GPT-4 Distillation #Multi-round Sliding Window

Downloads 55

Release Time : 12/18/2024

Model Overview

This model generates full-ranking lists as labels through a multi-round sliding window approach and is optimized with importance-aware training loss to enhance the ranking performance of large language models in long-context scenarios.

Model Features

Full Ranking Capability

Re-ranks 100 passages in one go without requiring sliding window strategies.

Multi-round Sliding Window Method

Generates full-ranking lists as training labels through multi-round sliding windows.

Importance-aware Training Loss

Utilizes a specially designed training loss function to optimize ranking performance.

Model Capabilities

Passage Re-ranking

Long-context Processing

Text Relevance Evaluation

Use Cases

Information Retrieval

Search Engine Result Ranking

Ranks multiple passages returned by a search engine based on relevance.

Demonstrates excellent performance on the BEIR benchmark.

🚀 RankMistral100

A full ranking model distilled from GPT - 4o, aiming to directly rerank 100 passages at a time.

🚀 Quick Start

This README provides detailed information about the RankMistral100 model, including its release background, training framework, backbone model, and performance.

✨ Features

Full Ranking Ability: The model can directly rerank 100 passages at a time, abandoning the sliding window strategy.
Innovative Training Approach: A multi - pass sliding window approach is proposed for generating the full ranking list as a label, and an importance - aware training loss is designed for optimization.
High Performance: Surpasses the strong baseline RankZephyr by 1.2 points on BEIR Avg.

📚 Documentation

Model Information

We release the full ranking model RankMistral100 distilled from GPT - 4o - 2024 - 08 - 06 used in Sliding Windows Are Not the End: Exploring Full Ranking with Long - Context Large Language Models.

Useful links: 📝 Paper • 🤗 Dataset • 🧩 Github

Training framework

Our full ranking model aims to directly rerank 100 passages at a time, abandoning the sliding window strategy. We propose a multi - pass sliding window approach for generating the full ranking list as label and design an importance - aware training loss for optimization.

The image

Backbone Model

RankMistral100 is finetuned from https://huggingface.co/mistralai/Mistral - 7B - Instruct - v0.3.

Performance

We surpass the strong baseline RankZephyr with 1.2 points on BEIR Avg.

Models	Covid	DBPedia	SciFact	NFCorpus	Signal	Robust04	Touche	News	Avg.
BM25	59.47	31.80	67.89	33.75	33.04	40.70	44.22	39.52	43.80
monoBERT (340M)	73.45	41.69	62.22	34.92	30.63	44.21	30.26	47.03	45.55
monoT5 (220M)	75.94	42.43	65.07	35.42	31.20	44.15	30.35	46.98	46.44
RankVicuna (7B)	79.19	44.51	70.67	34.51	34.24	48.33	33.00	47.15	48.95
RankZepeyer (7B)	82.92	44.42	75.42	38.26	31.41	53.73	30.22	52.80	51.15
RankMistral₁₀₀ (7B)	82.24	43.54	77.04	39.14	33.99	57.91	34.63	50.59	52.40

📄 License

This model is released under the MIT license.

💡 Usage Tip

If you use this model, please star our GitHub repository to support us. Your star means a lot!

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご