feat(ascend): add `flash_attention` operator#783

Draft

zhangyue207 wants to merge 1 commit into

InfiniTensor:masterfrom

zhangyue207:feat/ascend-flash-attention-operator

zhangyue207 commented Jun 30, 2026 •

edited

Loading

Collaborator

Summary

Add the Ascend flash_attention implementation.
Keep the existing upstream base operator API and drop the origin-only sliding_window extension from this split.
Add tests/test_flash_attention.py coverage.

Validation

git diff --check upstream/master...HEAD
clang-format --dry-run --Werror on changed C++ files
ruff format --check tests/test_flash_attention.py
ruff check tests/test_flash_attention.py


          feat(ascend): add flash_attention operator

6069ef8

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet