The issue seems to be that when the input to your sigmoid implementation is negative, the argument to torch.exp becomes very large, ...
確定! 回上一頁