Hi folks,
I am looking to contribute towards the CUDA module. I see a lot of issues that could use some help. I have skills in CUDA and I am a contributor to quite a few open source repositories. Some of those include cub, thrust, etc.
So with that being said - could someone point me towards an issue that I could work on? The problem is that I only have a 3080, I do not have access to work station GPUs. So some issues that involve multi GPUs aren’t feasible for me to work on atm. I am looking through the issues tracker and unfortunately a lot of them would be hard to reproduce on my machine. I don’t mind solving bugs either - I’m sure that would be most appreciated 
Please let me know if you have any recommendations for cloud providers that are suitable for individuals
Thanks
You could try to filter for good starter issues on GitHub by using label:"good first issue" label:"module: cuda" which should point to a few issues. Once you’ve found an interesting one, leave a comment in the issue explaining you are interested to work on it to avoid duplication of work.
In case you haven’t seen it yet, the contribution guide is also helpful.
Perhaps I should’ve picked an easier first issue considering this is my first contribution(and I’ve never really used Pytorch before tbh), but I did push this
Include support for the scatter gather cuda kernels to allow for comp… by ZelboK · Pull Request #124809 · pytorch/pytorch (github.com)
Please let me know if I need to do anything
1 Like
So sorry to bug again! Looking through the open issues, I’ve encountered quite a few that could be closed, some that are mostly users getting clarification, and others that are already getting worked on.
Could you point me towards any issue if possible? I don’t mind a more difficult task so long as I can reproduce the problem in my environment easily I also have a little bit of experience with CUTLASS for example if relevant.
I see that @janeyx99 already responded in the GitHub issue, but let me know if you cannot find anything using the query.
Sorry for not getting back earlier, was busy with work. I don’t get to contribute during work hours unfortunately🥲. I actually found one and pushed a PR up(I still think it needs some work, but before I proceed I’d like to get some reviews).
If there are any tasks that you think would be useful/impactful solved sooner than later please lmk.
Allow linalg.lstsq to use svd to compute the result for rank deficient matrices. by ZelboK · Pull Request #125110 · pytorch/pytorch (github.com)
^ is the second PR. My first one needs to be looked by a core reviewer or maintainer before it gets merged seems like.