## 🐛 Bug `t.sum(torch.float)` is broken yet `t.sum(dtype=dtype)` works. The error message is confusing as well. 