Li Junjie

找了半天没找到怎么整

It should be Fd = − log||∇D − Ex[log D(x)] − Ez[log(1 − D(G(z)))]||. But in terms of code, it's more like implementing Fd = log||∇D||. I think it should...

The code uses the `all_gather_with_grad` function to collect the tensor and gradient on all GPUs in order to compute the comparison loss across GPUs. I can successfully train the BLIP-2...