Results
3
issues of
shhhz
现公式没有很好地说明为什么每个 o_k 按常数进行的移动不会改变softmax的返回值： 建议修改为：
Calling the `demo2()` will result in an error, as numpy objects is incorrectly used to invoke the PyTorch method. Seems to be a problem introduced during the PyTorch refactoring.
Impressive examples! Does it support SDXL?