Academic or research source. Check the methodology, sample size, and whether it's been replicated.
Use version check for XPU fallback registration in inductor
Use version check for XPU fallback registration in inductor Replace torch.xpu.is_available() with torch.xpu._is_compiled() in: * Inductor fallback registration for embedding_dense_backward *...
PyTorch Releases··README: ~3 min
2-Minute Brief
According to PyTorch Releases: Use version check for XPU fallback registration in inductor Replace torch.xpu.is_available() with torch.xpu._is_compiled() in: * Inductor fallback registration for embedding_dense_backward * Inductor bmm dtype kernel selection to avoid unnecessary device queries during PyTorch initialization. Fixes #174672 Testing: Existing XPU fallback tests cover this path No behavioral change - same fallback registered Pull Request resolved: #174679 Approved by: https://github.com/etaf , https://github.com/Ei
Use version check for XPU fallback registration in inductor
TLDR
Use version check for XPU fallback registration in inductor Replace torch.xpu.is_available() with torch.xpu._is_compiled() in: * Inductor fallback registration for embedding_dense_backward *...
According to PyTorch Releases: Use version check for XPU fallback registration in inductor Replace torch.xpu.is_available() with torch.xpu._is_compiled() in: * Inductor fallback registration for embedding_dense_backward * Inductor bmm dtype kernel selection to avoid unnecessary device queries during PyTorch initialization. Fixes #174672 Testing: Existing XPU fallback tests cover this path No behavioral change - same fallback registered Pull Request resolved: #174679 Approved by: https://github.com/etaf , https://github.com/Ei