unexpected `inference_mode` interaction with `torch.autograd.functional.jacobian` about pytorch HOT 3 OPEN

I'm not sure if this is the intended behavior

This is expected behavior. Computing the jacobian with torch.autograd.functional.jacobian does not work with inference mode because it needs to record computation to the autograd graph underneath.

perhaps we can add a line in the documentation of inference mode about this behavior?

From the docs there is:

InferenceMode is a new context manager analogous to no_grad to be used when you are certain your operations will have no interactions with autograd

From https://pytorch.org/docs/stable/generated/torch.autograd.grad_mode.inference_mode.html

Does this line make things clearer?

from pytorch.

I see, yes that certainly makes things clearer. Maybe we can add something like that to the documentation at https://pytorch.org/docs/stable/notes/autograd.html#inference-mode? So maybe instead of

Enable inference mode when you are performing computations that don’t need to be recorded in the backward graph, AND you don’t plan on using the tensors created in inference mode in any computation that is to be recorded by autograd later.

we can write

Enable inference mode when you are performing computations that do not have interactions with autograd, AND you don’t plan on using the tensors created in inference mode in any computation that is to be recorded by autograd later.

from pytorch.

Sounds OK to me, thanks for the suggestion. Happy to accept a PR with that change.

from pytorch.