Academic or research source. Check the methodology, sample size, and whether it's been replicated.
: Factor out scaled_mm algo checks to non-CUDA ()
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native .
PyTorch Releases··README: ~3 min
2-Minute Brief
Major AI lab announcement.
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native .
Part of a refactor of #170670 Test Plan: Functionality-neutral.
Open receipts to verify and go deeper.
8-Minute Deep Dive
Context
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native . Part of a refactor of #170670 Test Plan: Functionality-neutral. pytest test/testscaledmatmulcuda.py Co-authored-by: @CWOA Signed-off-by: Simon Layton simonlayton@meta.com Pull Request resolved: #175657 Approved by: https://github.com/drisspg
For builders
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native .
Verify
Prefer primary announcements, papers, repos, and changelogs over reposts.
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native .
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native .
Part of a refactor of #170670 Test Plan: Functionality-neutral.
Open receipts to verify and go deeper.
8-Minute Deep Dive
Context
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native . Part of a refactor of #170670 Test Plan: Functionality-neutral. pytest test/testscaledmatmulcuda.py Co-authored-by: @CWOA Signed-off-by: Simon Layton simonlayton@meta.com Pull Request resolved: #175657 Approved by: https://github.com/drisspg
For builders
Summary: In preparation for an XPU-specific backend for scaledmmv2 , move some helpful routines out of CUDA-specific code and into general aten::native .
Verify
Prefer primary announcements, papers, repos, and changelogs over reposts.