Unable to train H3M model after updating torch for cuda compatibility

[anupampani]

Hey!
Great project and I want to re-run the code to get some more insight. However, after updating pytorch (and a couple of other dependencies), I was able to obtain the padded features but when I try to train the H3M model, I run into a
" TypeError: expected Tensor as element 0 in argument 0, but got NoneType

[03/01 17:26:25][DEBUG] ego4d.datasets.ptv_dataset_helper_features: 113: Failed to load feature with error: [Errno 2] No such file or directory: '/mnt/lv1/ego4d/icvae/v2/features_pad/cf0e2eca-2527-4673-8e44-02cd620590be_121.pt'; trial 0"

Presumably because of a lot of failed load features. I was wondering if anyone else has had this problem while re-creating