onnxruntime
ebac9e56 - Enable grouped INT8 sparse attention for four-row verification

Commit
8 days ago
Enable grouped INT8 sparse attention for four-row verification Add an opt-in maximum-row gate while preserving the two-row default, and extend grouped/vector-staging reference shapes through four rows. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Author
Parents
Loading