ZZK

第一点，我们的amp暂时在Graph层支持（即静态图），Eager动态图还不支持AMP，所以直接复制amp模块import我大概率估计是跑不起来

> Hi, > > I am trying the stable diffusion in the example https://github.com/facebookincubator/AITemplate/tree/main/examples/05_stable_diffusion > > But get the following error when compiling the model with python3 examples/05_stable_diffusion/compile.py --token ACCESS_TOKEN...

> > Can you try this commit [8c158f5](https://github.com/facebookincubator/AITemplate/commit/8c158f54278975a074f9cadf12982bcba259436d), it works well on my side. > > Tried out yours after the current master and all other versions failed to compile....

> > Hi @sleepcoo, Is the bug fixed now? We will add the code format checker later. 🙏 Could you wrap up this PR first? > > I fixed it,...

If I use this sin activate in CNN to some simple task, such as Image Classification. I just follow the initialization in your code not the paper right?I was confused...

> > > 这样会造成 s->b 的转化 > > > > > > s->b 会带来什么问题呢 0。0 > > 会产生一些 unexpected 的行为，比如 loss 会变成 b，本来 loss 应该是 P 的 :< 感觉你可以把计算loss的那段代码贴一下，应该更清楚点

他这里是自定义了gelu吗，因为我们单测是有和torch版本的gelu比较的。 如果可以的话，用torch.nn.GELU再看看结果

方便贴下你的结果么，我这里mac cpu，pytorch1.7.1是 ```python import torch import numpy as np x = torch.Tensor(np.array([-5, 10, 105])) gelu = torch.nn.GELU() out = gelu(x) tensor([-1.4333e-06, 1.0000e+01, 1.0500e+02]) ```

嗯嗯，这确实是个很细节的点，之前我们做的可能都没有注意到这个细微的差别