You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
My team(me, @jinotter3, @suzyvaque) is proposing the integration of the executorch tensor filter into nnstreamer. Supporting executorch could enhance nnstreamer's capabilities by enabling efficient execution of PyTorch models directly.
Currently, nnstreamer supports various tensor filters but lacks integration with executorch. Including this could not only expand nnstreamer's usability but also attract more PyTorch users to nnstreamer.
Thank you for considering this enhancement. I look forward to your feedback and any suggestions you might have.
Please write in a single C++ file, inheriting tensor_filter_subplugin class. For example, tensor_filter_snpe.cc.
Please make it portable. It should be able to be built (and tested) in a new Ubuntu machine without any (unscripted) changes. Ultimately, it should be able to be built by pdebuild (or PPA) if you consider Ubuntu as the standard environment.
You may be based on the docker file (/tools/docker). If you can build executorch subplugin in the docker by adding one or two lines that "installs" executorch in the docker, you can ensure its portability.
Or, in other words, you need to guarantee that anyone else in the world can get the same build results without reading your build manual, special build scripts, or description.
PyTorch is releasing ExecuTorch for edge device inferences.
The text was updated successfully, but these errors were encountered: