Closed Pourbahman closed 2 years ago
(Most) results are on electra-base-msmarco, unless you mean v1 of the arxiv paper which primarily used bert-base-msmarco.
The code here was used to train on smaller models, but this should work fine in capreolus with minor changes as well.
Hi Andrew,
According to https://github.com/capreolus-ir/capreolus/blob/3521171ecf38cebfec5e19e22621bf9dfabf58d9/capreolus/reranker/parade.py#L123 I found that you did not train your model on a smaller models like tiny BERT. Would you please tell me if I am right?
Also, according to my understanding the best values on your evaluation metrics is for bert-base-msmarco. Am I right?
Thanks in advance, Kind Regards