fengqing.lu de92296bbb [Intel GPU] undo broadcast on zero stride tensor for SDPA (#151976) ...

Fix https://github.com/pytorch/pytorch/issues/152290.

The model **hubert** uses aten::expand to build attention mask by broadcasting. Pytorch uses strides[d]=0 to represent broadcast, which is not supported by oneDNN.  This PR handles this scenario.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/151976
Approved by: https://github.com/guangyey, https://github.com/EikanWang, https://github.com/drisspg

2025-05-14 16:09:03 +00:00

..

conda

PyTorch -> C++17 (#98209) (#100557)

2023-05-19 00:49:08 +00:00

src

[Intel GPU] undo broadcast on zero stride tensor for SDPA (#151976)

2025-05-14 16:09:03 +00:00

tools

[BE][CI] Get rid of duplicated code (#131406)

2024-07-23 04:01:13 +00:00

CMakeLists.txt

Revert "Remove THC from OSS build (#134969)"

2024-11-26 00:00:12 +00:00