Training Parameters for BERT classification on MSMarco

[gsgoncalves]

Hi,

Congrats on your project!

Could you please share the training details that lead to the BERT models finetuned on MSMarco. I am especially interested in the learning rate, batch size, and the number of steps.

Did you follow all the specifications given by Nogueira and Cho? I could see that this is reference number 2 in your benchmarks section. However, the text that refers to this reference seems to be missing.

Thank you

[davidmrau]

I'd be also interested in that!