fix: Use `is_tensor_like` instead of `instance(Tensor, · )`#583

Merged

PierreQuinton merged 5 commits intomainfrom

use-is-tensor-like

Feb 19, 2026

Commits on Feb 19, 2026

Fix in `_jacobian_computer`. The rational here is to let torch fail if the operations where not implemented but allow custom implementations.
PierreQuinton
committed
Change in `module_hook_manager`. I think this one is not safe as the `require_grad` field might not exist in something that is not a `Tensor`. I'm not sure the status of this in torch so we should …
PierreQuinton
committed
This one makes sense, but we are forced to cast as they did not make `is_tensor_like` a `TypeGuard`. I opened an issue: https://github.com/pytorch/pytorch/issues/175324
PierreQuinton
committed
Same in `jac`, need cast but makes sense.
PierreQuinton
committed
Same with `autojac._utils`
PierreQuinton
committed