Mario Lezcano Casado
Mario Lezcano Casado
@ThomasRaoux addressed the review (at long last)
No need really. Decompositions are there to teach `torch.compile` how to trace through that function. Just make sure you add the function name to the relevant expect file that says...
I'm not going to be able to continue reviewing this PR. Could you take over @peterbell10 for the next week? After that, perhaps @amjames can take over.
can you rollback the pin in the submodules :) (it happens to the best of us)
Also, the vjpvjp tests, if they fail, you should probably just xfail them in the relevant file, as we are not implementing second derivatives here. You could perhaps implement the...
@albanD this PR is in a pretty good state already, could you find someone that can shepherd it?
Hi Pablo! I'm afraid I'm going to need a small reproducer where you hit this issue. For debugging, consider using https://pytorch.org/docs/stable/autograd.html#torch.autograd.detect_anomaly to debug this sort of issues.
It's seems then that it was fixed in newer PyTorch and, at the end of the day, PyTorch 1.12 is ancient so you should upgrade. Closing then!
Oh, the first external contribution to the project! Sadly, the project is a bit out of date, but given that it's useful for people, I'll try to spend some time...
tensorial_size is not supported since the resolution of the vectorized Lyapunov equation is annoying the kroenecker product I am not sure how this is related? `tensorial_size` is just the batched...