Question about model architecture

[Lim-Sung-Jun]

Hello,

I'm trying to reimplement the model and I found that the situation between train and inference is different. In the train.py, the model feed the input like format as (query, doc) but in the precompute.py, the model feed input as (query) and (doc) seperately. According to your paper, embed(query) and embed(doc) is right I think. Am i miss something?

Thank you