Re: GSOC 2024: Expanding ROCm support within Debian and derivatives
Hi Sanket!
On 2024-03-04 13:04, Sanket Sharma wrote:
> I was looking at the GSOC 2024 page and "Expanding ROCm support within
> Debian and derivatives" caught my attention.
>
> I'm not a student and I've participated in GSOC once before way back in
> 2006. I have some free time now and would like to get back into open
> source. I have worked extensively as DevOps and Data Engineer with some
> exposure to machine learning, but not a lot. Recently, I was involved in
> a project where I had to upgrade a really old system that used x64
> versions of tensorflow. The libraries were from 2018 or so (Tensorflow
> 2.2.2 ?) and were not available on M1/ARM by default. The project
> involved learning Bazel to build Tensorflow, python wheels, publishing
> etc. figuring out build dependencies,incompatibilities, running tests etc.
> While I haven't worked directly with CUDA or ROCm and it's quite
> different to what I did, I'm keen to learn and will be very happy to
> join the project long term even beyond GSOC.
>
> Is there any documentation available that I can refer to before I make a
> formal proposal for the project? I've been busy going through the
> contributor guidelines but couldn't find anything specific to this
> project so far.
>
> Any help, guidance and or tips would be greatly appreciated.
I really hate to say this, but given your experience, I'm not sure you
qualify as a GSoC applicant. The FAQs [1] state "GSoC is not intended
for experienced software engineering professionals".
However: we're happy to welcome any new contributors, GSoC or not. GSoC
was meant to be an enticement. If you're generally interested in
contributing to our team (or any other parts of Debian), you can do so
at any time and for as long as you wish.
Our current work is focused on getting the ROCm stack, and software that
can make use of it, packaged for Debian and Ubuntu. In this process,
we're also setting up our own infrastructure, for example a CI
environment that can make use of GPUs. We're building our own tooling
for this, so this could be up your DevOps ally.
But the debian-ai list goes beyond that, and there are other groups
(like the Debian Science Team) that are also always happy to welcome new
contributors.
If that is something that interests you, you can follow up here anytime.
If, however, GSoC is a must, I'd suggest that you first check with GSoC
to confirm that you qualify (it's their call, not ours).
Best,
Christian
[1] https://developers.google.com/open-source/gsoc/faq
[2] https://ci.rocm.debian.net/
Reply to: