UKPLab/gpl

Powerful unsupervised domain adaptation method for dense retrieval. Requires only unlabeled corpus and yields massive improvement: "GPL: Generative Pseudo Labeling for Unsupervised Domain Adaptation of Dense Retrieval" https://arxiv.org/abs/2112.07577

57
/ 100
Established

Generated queries are synthesized from the corpus using T5 models, then negative examples are mined via dense retrievers, and pseudo-labels are assigned by cross-encoders to create training signal without manual annotation. Built on Hugging Face Transformers and Sentence-Transformers, GPL accepts BeIR-format datasets and trains dense retrievers using MarginMSE loss optimized for dot-product similarity.

340 stars and 175 monthly downloads. No commits in the last 6 months. Available on PyPI.

Stale 6m
Maintenance 0 / 25
Adoption 15 / 25
Maturity 25 / 25
Community 17 / 25

How are scores calculated?

Stars

340

Forks

38

Language

Python

License

Apache-2.0

Last pushed

Jul 06, 2023

Monthly downloads

175

Commits (30d)

0

Dependencies

4

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/UKPLab/gpl"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.