On 5/19/20 4:04 PM, Mo Zhou wrote: > Package: lists.debian.org > Severity: wishlist > X-Debbugs-CC: debian-de...@lists.debian.org > > (I think I should CC -devel since I'm requesting a NEW dev ML) > > Hi list master, > > I'd like to request for a new public, and archived mailing list > > Name: debian-mlhwaccl > Short-Desc: "Debian's Machine Learning & Hardware Acceleration" > Long-Desc: Development of Machine/Deep Learning Software and > Corresponding Hardware Acceleration Measures incl. ROCm, OpenCL, SYCL > and CUDA > Category: obviously a dev list > Subscription-Policy: open > Post-Policy: open > Web-Archive: yes (we definitely need an archive) > > > Our demand / Rationale: > > 1. we need an archived mailing list to better coordinate the works > related to deep learning / machine learning and hardware acceleration > (including ROCm, CUDA, OpenCL, SYCL) specific to machine learning > applications. Debian science team maling address is not quite > appropriate for the topics. > > 2. we need an archived mailing list as the maintainer address for the > ROCm team to better coordinate the works. We should not abuse debian > science team's alioth mailing list address as the maintainer address > neither. > > 3. we need an archived mailing list for discussing about the > development of debian deep learning team. The machine learning > and deep learing subareas are already significant enough to form > a dedicated interest group inside debian. > > Teams involved: > Debian ROCm Team, Debian Deep Learning Team > > Expected key words for the mailing list: > tensorflow, pytorch, bazel, sklearn, opencl, sycl, rocm, cuda, llvm, > xla (tensorflow), openmpi > > as you see, some times the topics could be off-topic to debian-science, > and the "machine learning & hardware acceleration" things always > involves complicated intersection of various areas. It could be better > if we have a dedicated list for the whole thing!
there's also the offloading support in GCC, already in the archive as gcc-10-offload-nvptx and gcc-10-offload-amdgcn, maybe not as much used as the llvm based projects. Matthias
