Question about torch.lstsq and torch.linalg.lstsq

[Kidleyh]

My pytorch version is 1.12.1.
I used torch.lstsq(A, B)[0][B.shape[1]] and torch.linalg.lstsq(B, A).solution and the results are different. I want to get the answer of the former. How should I use torch.linalg.lstsq to get it?

cc @jianyuh @nikitaved @pearu @mruberry @walterddr @xwang233 @lezcano

[lezcano]

Can you update to the lastest release? Chances are that the bug is already fixed in the lastest release.

If it has not, can you provide a self-contained reproducer of the issues?

Also, note that the fact that even if the returned matrices are different, X = linalg.lstsq(A, B) just guarantees that torch.allclose(A @ X, B). There may be many matrices X that may solve this equation though.