Repository for the paper "Continual Learning of Long Topic Sequences in Neural Information Retrieval". In this repository you will find the code for generating the long sequences topics/tasks.

pytorch : conda install pytorch torchvision torchaudio pytorch-cuda=11.6 -c pytorch -c nvidia

Transformers and pyserini pip install transformers pyserini numpy

Dataset can be generated based on their topics using a sentence bert pretrained model. We also provide the three set used in the associated paper.

In the following we describe the different approaches tested in successively fine-tunning model.