Skip to content
Branch: master
Find file History
Latest commit 5d4d58b Mar 20, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
Duet.ipynb Rename Duet-PyTorch-WordEmb-ReLU.ipynb to Duet.ipynb Feb 20, 2019
req.txt updating duet Nov 20, 2018

Duet Baseline

To help the community reproduce our results and get started quickly, we are including a Duet v2 baseline on our leaderboard. The details of this model can be found in the following paper:

For the Duet v1 model, please see: and


Python 3.5, CUDA 9.0, Pytorch, numpy, jupyter


The Duet v2 code is mostly self-contained. It requires two extra data files, that are not part of the official MS MARCO dataset but are included with the code.

Vocabulary file

The vocabulary file contains terms from the passage collection whose document frequency is above a certain threshold. The code references "word-vocab-small.tsv" which contains 71,486 terms from the MS MARCO passage collection.

Inverse document frequency file

This file contains terms and their corresponding inverse document frequency (IDF) as computed according to the specification in the paper.

Important: The file "idfnew.norm.tsv" only contains terms from queries in MS MARCO train/dev/eval sets. This is because Duet v2 only needs IDF values corresponding to the terms in the query. This is not a general purpose IDF dataset for MS MARCO. If you need one, please make sure to generate one directly from the MS MARCO data.


Please cite the following papers, if you use the Duet model in your research.

Duet v2

  title={An Updated Duet Model for Passage Re-ranking},
  author={Mitra, Bhaskar and Craswell, Nick},
  journal={arXiv preprint arXiv:1903.07666},

Duet v1

  title={Learning to match using local and distributed representations of text for web search},
  author={Mitra, Bhaskar and Diaz, Fernando and Craswell, Nick},
  booktitle={Proceedings of the 26th International Conference on World Wide Web},
  organization={International World Wide Web Conferences Steering Committee}

Duet v1 on TREC CAR

  title={Benchmark for complex answer retrieval},
  author={Nanni, Federico and Mitra, Bhaskar and Magnusson, Matt and Dietz, Laura},
  booktitle={Proceedings of the ACM SIGIR international conference on theory of information retrieval},
You can’t perform that action at this time.