LukeLin

we need use some codes like: ``` device = torch.device("cuda:{}".format(rank)) model = Net().to(device) data, target = data.to(device), target.to(device) ```

line120, should change to `epoch_loss += loss.data.item()`

I realize it because I have two .yaml to control the icons, two .yaml will add up.

> _No description provided._ e_social: enable: true e_social_links: Google Scholar: / _**you can add your link in this line**_|| fas fa-graduation-cap ORCID: / || fab fa-orcid

I realize that this theme should ban the command `hexo clean` because it will delete the dir /img , /attach and so on.

Many workers accumulate variables may lead to out of memory? I guess.

as well as in graph_sage_unsup_quiver.py

I tried uninstall and install 0.1.1. It will hang at: https://github.com/quiver-team/torch-quiver/blob/0592669225954e46d9c57c0ca121c365af47b57f/examples/pyg/reddit_quiver.py#L29 as well as https://github.com/quiver-team/torch-quiver/blob/0592669225954e46d9c57c0ca121c365af47b57f/examples/pyg/ogbn_products_sage_quiver.py#L41 use pdb to find problem. it hang at srcs/python/quiver/pyg/sage_sampler.py self.quiver = qv.device_quiver_from_csr_array(self.csr_topo.indptr, self.csr_topo.indices, edge_id, device,...

> Thanks for the ticket! https://github.com/quiver-team/torch-quiver/blob/main/srcs/cpp/src/quiver/cuda/quiver_sample.cu#L361-L461 The function calls this C++ function internally, and in your case it is hanging somewhere the middle. May you know how can i sort...

I figure it out, because I use a A100 GPU. compiled codes cannot run at A100. I change to V100 and run successfully.