Something wrong with affine coupling?

[SiyuWang15]

When I set use affine coupling rather than additive coupling, training on cifar10, n_flow = 32, n_block = 3, the weights will rapidly become Nan, and sample s become all black. But the same phenomenon won't appear when I use additive coupling.

[weepiess]

When I set use affine coupling rather than additive coupling, training on cifar10, n_flow = 32, n_block = 3, the weights will rapidly become Nan, and sample s become all black. But the same phenomenon won't appear when I use additive coupling.

Hi, I met the same problem. Have you fixed it yet?

[Weijiang-Xiong]

same problem here, did you find a fix or a walk around?

[birdortyedi]

same problem for me, any update? thanks.

[sidney1505]

I still experience the same problem