Hi! Hope all is well. Thanks again for sharing the great job!
I've been following your work recently and I used THOP to test MACs and PARAMs of the network. The results are shown in the figure below, where PARAMS is consistent with the paper, while MACs is only 146.63G. Considering that flops are roughly twice as large as MACs, there is still a certain gap. If I adjust the input batch_size from 1 into 4, MACs will expand four times, which is 586.52G. So I'm wondering if there is something wrong with my experiment that makes MACs different from the one reported in the paper. And it will be helpful to know how you test the FLOPs!
Hi! Hope all is well. Thanks again for sharing the great job! I've been following your work recently and I used THOP to test MACs and PARAMs of the network. The results are shown in the figure below, where PARAMS is consistent with the paper, while MACs is only 146.63G. Considering that flops are roughly twice as large as MACs, there is still a certain gap. If I adjust the input batch_size from 1 into 4, MACs will expand four times, which is 586.52G. So I'm wondering if there is something wrong with my experiment that makes MACs different from the one reported in the paper. And it will be helpful to know how you test the FLOPs!
Thank you very much!