Finetuning BERT for Information Retrieval

This is the code base for NTU-AI6103 Deep Learning & Application Course. We explore different Bert-based model design for dealing with different different downstream task in Information Retrieval.

We developed based on the neural-ir repo. In addition to adjusting the bi-encoder and cross-encoder architecture models for training, we also re-built ColBertv2 based on the ideas in the paper and performed training based on the knowledge distillation.