upgrading to the latest version of Spacy

Upgrading Spacy, b/c old versions often won't install properly via pip

new models + various fixes

This version improves in a lot of tiny details and it has the following major features:

1. Supporting embeddings using any model from the Sentence BERT library.
2. Codeless dataset creation for (almost) any dataset in the IR-datasets library.
3. Improved re-ranking model zoo, including:

• COLBERT (in the re-ranking mode)
• Any sentence BERT bi-encoder
• PARADE models & Longformer

pypi version

This is our first pypi-installable version of the package.

Reproducing traditional run on the MS MARCO leaderboard

This short documents describes a traditional IR system that achieved MRR@100 equal to 0.298 on the MS MARCO leaderboard (on 2020-12-06). Although inferior to most BERT-based models, it outperformed five neural submissions,
including two submissions that used a large pretrained Transformer model for re-ranking. We provide software and data to reproduce our results.