This model has been pretrained on BEIR corpus then finetuned on MS MARCO with BM25 warmup only, following the approach described in the paper
COCO-DR: Combating Distribution Shifts in Zero-Shot Dense Retrieval with Contrastive and Distributionally Robust Learning. The associated GitHub repository is available here
https://github.com/OpenMatch/COCO-DR.
This model is trained with BERT-large as the backbone with 335M hyperparameters.
