Add ExportFriendlyMultiheadAttention for dynamic shape torch.export#2

Closed

rbavery wants to merge 3 commits into

mainfrom

ryan/export-friendly-mha

Commits on Jan 22, 2026

fix: enable CPU export and inference by removing pin_memory on scale tensor
rbavery
committed
Add ExportFriendlyMultiheadAttention for dynamic shape torch.export
rbavery
committed
Remove H/W keyed caches for torch.export dynamic shape support
rbavery
committed