Projects

Fast-Forward Indexes

Fast-Forward indexes combine the efficiency of lexical sparse retrieval methods with the effectiveness of semantic dense retrieval methods. The main features are:

• Fast CPU-based ranking
• Memory- and disk-based indexes
• Index compression via sequential coalescing
• Heuristic early stopping technique to limit the number of index look-ups
• PyTerrier integration

pip install fast-forward-indexes

Documentation | GitHub | PyPI

Ranking Models

This repository contains the implementations of our ranking models as well as some baselines. It supports training, validation, testing, and re-ranking.

GitHub

Ranking Utilities

This library provides utilities for training ranking models, offering integration with PyTorch Lightning.

• Dataset parsing and pre-processing
• Pointwise and pairwise training loss
• Validation using ranking metrics

GitHub