InRanker is a version of monoT5 self-distilled from monoT5-3B with increased effectiveness on out-of-domain scenarios. Our key insight were to use language models and rerankers to generate as much as possible synthetic "in-domain" training data, i.e., data that closely resembles the data that will be seen at retrieval time. The pipeline used for training consists of two distillation phases that do not require additional user queries or manual annotations: (1) training on existing supervised soft teacher labels, and (2) training on teacher soft labels for synthetic queries generated using a large language model.

The paper with further details can be found here. The code and library are available at https://github.com/unicamp-dl/InRanker

The library was tested using python 3.10 and is installed with:

The code for inference is:

@misc{laitz2024inranker,
      title={InRanker: Distilled Rankers for Zero-shot Information Retrieval}, 
      author={Thiago Laitz and Konstantinos Papakostas and Roberto Lotufo and Rodrigo Nogueira},
      year={2024},
      eprint={2401.06910},
      archivePrefix={arXiv},
      primaryClass={cs.IR}
}