f-dangel / f-dangel/backpack

Resnet + DiagHessian

Open
#344 1 comment 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
617
Forks
57
PR merge metrics
No merged PRs in 30d

Description

Hi!

I trained a ResNet18 from torchvision.models on my dataset (CIFAR-10) and now I would like to compute the Exact DiagHessian but I'm encountering the following error:

NotImplementedError: Extension saving to diag_h does not have an extension for Module

I followed all the instructions as per tutorial so I extended the model in evaluation mode to avoid the batch normalization layers.

If I try with other functions like DiagGNNExact it works but is very slow, while DiagHessian returns me this error.

The problem is that even if I avoid the resnet from torchvision following the tutorial on residual networks it made use of SumModule.

Thank you!

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.