functorch

functorch copied to clipboard

Get wrong jacobian from copyslice operation

8

Hi, Thanks for your great work, I'm working with some minimize algorithm accleration using this code and find a problem. Here is a sample example to reproduce the conditon: ```python...

KagamineLenOffical

[hotfix] fix build error from FuncTorch dispatch key

2

Reviewed By: ezyang Differential Revision: D39374319

winnie1994

memory_efficient_fusion leads to RuntimeError for higher-order gradients calculation. RuntimeError: You are attempting to call Tensor.requires_grad_()

3

Hi All, I've tried improving the speed of my code via using `memory_efficient_fusion`, however, it leads to `Tensor.requires_grad_()` error and I have no idea why. The error is as follows,...

AlphaBetaGamma96

Simultaneous computation of per-sample gradient and per-batch gradient

1

Hi. Thanks for providing functorch, one of most effective lib. Actually i am working on my code which needs an access into both per-sample gradient in mini-batch and per-batch gradient,...

SJShin-AI

No differentiated Tensors in the graph when using autograd.grad with functorch

1

Hi all, I have some problem when computing the vjp (with torch.autograd.grad) between the jacobian of a neural network w.r.t. its parameters and a random tensor. The neural network evaluates...

malessandro

Per-sample-gradient: Get gradient 0 when using grad(params_tograd, params) with respect to part of model's parameters

1

Hi Pytorch team, recently I need to calculate per sample's gradient with respect to part of model's parameters. The problem is that for the toy example, it works. But for...

Ancientshi

Will pmap be supported in functorh？

2

Greetings, I am very grateful that vmap is supported in functorch. Is there any plan to include support for pmap in the future? Thank you. Additionally, what are the ways...

shixun404

wrapper->level().value() <= current_level INTERNAL ASSERT FAILED at "../aten/src/ATen/functorch/ADInterpreters.cpp":39

1

I met this issue when I use `jacrev` function to calculate the Jacobian matrix using a VisionTransformer. The exception occurs when I move the VisionTransformer to GPU. RuntimeError('wrapper->level().value() level().val.... escaped?',)...

ElegantLin

Add vmap support for PyTorch operators

2

We're looking for more motivated open-source developers to help build out functorch (and PyTorch, since functorch is now just a part of PyTorch). Below is a selection of good first...

zou3519

Support statically shaped dynamic operations (nonzero, unique, etc)

13

## 🚀 Feature Cross-post of https://github.com/pytorch/pytorch/issues/62320. torch.nonzero, torch.unique, etc should accept an optional size argument to produce statically shaped output. We should identify all the PyTorch operations that need this...

zou3519