Crush

> My thoughts on the Activity Viewer: > > ## What goes in the activity viewer? > * In the left: I think we can follow the design of XCode...

This would be a super nice feature, and can let people get main idea or valueble part of articles with a glance, as a PhD or researcher, this save lives...

> Thanks for reporting! What is the timing when only running a single sparse-sparse matrix multiplication (e.g., `adj_t @ s`)? My assumption is that the intermediate sparse matrix is pretty...

I leave two variables as .pkl here, you can try it: [test.zip](https://github.com/rusty1s/pytorch_sparse/files/9443195/test.zip)

> How dense is the final output? I think doing sparse-sparse matrix multiplication on GPUs efficiently is still an open problem. This is not really a GPU-friendly algorithm, as it...

> This is not really a GPU-friendly algorithm I also tried to do ot on CPU, and reduce time from ~110 to ~25:

> The memory scales with the number of non-zero entries. DiffPool will ultimately create a dense output adjacency matrix. It also looks like you are not treating batch dimensions correctly...

> > @TonyRL > > 你好，关于「非法访问」报错的问题，依旧存在 > > https://rsshub.app/bilibili/user/video/2267573 > > 我这里自建+设置UA+添加大量Cookie可以解决 > > 参见 #13365 请问具体到如果使用在.env中添加cookie的方式，要怎么操作呢？直接BILIBILI_COOKIE_xxx="cookie_1","cookie_2",...,"cookie_N"这样吗？好像不太对？

> > > > @TonyRL > > > > 你好，关于「非法访问」报错的问题，依旧存在 > > > > https://rsshub.app/bilibili/user/video/2267573 > > > > > > > > > 我这里自建+设置UA+添加大量Cookie可以解决 > > > 参见 #13365...

> > > > > > @TonyRL > > > > > > 你好，关于「非法访问」报错的问题，依旧存在 > > > > > > https://rsshub.app/bilibili/user/video/2267573 > > > > > > > > >...