Horace He

That's an interesting way to solve it, but the problem is that for a common pattern like ``` x = torch.randn(..., requires_grad=False) model(x).sum().backward() ``` this would result in the top-level...

We discussed this in triage review, and we do not believe that (modulo bugs), `torch.compile` should ever result in *worse* accuracy, particularly inference. So this sounds like a bug we'd...

You need to install Julia in order to generate the spectral embeddings. If you're too lazy, you can just run the models without the spectral embeddings - I think MLP...

Can you post the exact command you're running, as well as the stack trace?

Ah - comment out line 148 and 147 here: https://github.com/CUAI/CorrectAndSmooth/blob/master/gen_models.py#L148 to run without embeddings. To install Julia for generating the spectral embeddings, you need to install pyjulia in addition to...

Ah, that's a great point. For the most part, it should be PyTorch Geometric and Open Graph Benchmark. Deep Graph Library is needed for the GAT experiment. Those have their...

While I figure out how to present the dependencies in a nice way, here's the yaml for now: https://pastebin.com/3YrQqaBF Not all of them are necessary, and I'm not sure if...

We actually do currently have that for regular VSCodeVim, but it hasn't been ported over yet.

I think this issue should be solved with efac43773c7f6fbd19bb0500bd9abb737b819ab3 and e6e7d9d8d115cf6996a9c619b2af2e3d4a511fc7. Basically, how it works now is that the keybindings for package.json are now autogenerated through a script, and we...

If anybody thinks this isn't flexible enough, please suggest improvements! Keep in mind that you'll still be able to bind vim keybindings to call VSCode commands, but that will take...