The size of the lstsq() solution tensor is different from Normal equation

Let A is m x n tensor and B is m x k tensor.
Then size of normal equation tensor is n x k.
but the size of return tensor of torch.lstsq() is m x k.
Why does it works like this?