CX1888

I want to implement a dual branch model, where 2 backbones match with different dataset types. What should I do?

I trained the model for 3 days, still running! And why the loss is negetive?? One of the infos is as follows: 24-06-04 21:12:56.873 - INFO: loss: -8.6199e-01 l_amp: 1.2907e+09...