Samantha Andow

Moving on here since we started some internal discussion but it might get lost > Automatic sync with internal functorch. This seems difficult since nightly != internal master != what's...

Hi @yueyericardo, thanks for the issue! > what is the difference between using vmap(jacrev and composing vmap and torch.autograd.grad together to get jocabian? At a high level, `jacrev` "lives" in...

> Are there any suggestions for evaluating only part of the Jacobian and Hessian by using vmap and grad? Depending on what slice of the Jacobian or Hessian you want,...

Thanks @AlphaBetaGamma96 for linking to that, that's the right link for this. I wanted the error message to link to be more insightful and link to that so I'll investigate...

@zou3519 got it all! Only small things to add 😄 > Setting the model to eval() works for me This was something that was updated somewhat recently. I believe it...

Hi @sgstepaniants! This should be possible by doing `vmap(vmap(f, in_dims=(0, None)), in_dims=(None, 0))(x, v)` For a little more color, what `in_dims` is saying here is where the batch dimension of...

@vfdev-5 I am because I set it up for PyTorch development (this assumed that most people would be coming from PyTorch development so would already have it set up but...

@vfdev-5 Ahh good eye! I'm seeing that they actually recently removed the flake8 integration about a month ago. [This is the issue that talks about it](https://github.com/pytorch/pytorch/issues/69500). It looks like we...

@vfdev-5 I see! Sorry I misunderstood. I think setting that up is fine. Either using the script like above like described in the issue seems like it should be fine...

You guys totally have the right solution to drop from the right or bottom anything that's too big. Here's the reasoning I have for it and let me know if...