3 ms·
Nah it’s because PyTorch has a different implementation for __iadd__. It’s saving a copy by mutating the LHS in-place, and possibly more divergent as comments r
by masklinn 4y ago
Nah it’s because PyTorch has a different implementation for __iadd__. It’s saving a copy by mutating the LHS in-place, and possibly more divergent as comments report broken code.