[CPU] Implement X-Attention for intel CPU #32086
maxnick
merged
33 commits into
openvinotoolkit:master
from
mangguo321:mang/xatt
zhangYiIntel
commented
on 2025-10-16
Enable xattention for intel CPU
c9b96b5e
Integrate mask to PA
b4bbb28f
Support multiple sequences for xattention
4c435044
Fix find block issue
276fd0b1
Copy key buffer before executing GEMM and bf16 precision support.
a09b3a49
revert pa conflict changes with sparse attetnion execution
2408add3
enable sparse attetnion execution
a6ff6674
Use optimized softmax kernel in X-Attention
de6a6def
Optimize block sum using hardware instructions
715f790c
Add macro definications to not support ARM platform
a9c542c3
modify xattention_block_size and PA block_size check and scale calcul…
114ef7ff
optimize sparse softmax mask in sparse mask attention: extend attn_so…
c86732d9
Add const to unchanged variable
bcede76c
Minor changes in find blocks logic and remove useless parameter.
aac68ab6
Fix clang-format issue
ca4347d2
optimize use_softmax_sparse_mask check
c0ffb6af
Apply suggestions from code review
fcf896c4
Add test case.
8d390934
Fix Clang-format issue
bcb1309c
Avoid repeated brgemm kernel creation and temp buffer creation
a654802d
Fix ARM cross compile
1419bd60
Fix CI failure
feb4d228
Fix arm compile error
2ac85cbd
Fix RISCV64 error
a92433a7
Change sum_block logic to handle column num.
9f27b2f1
Add const to unchanged variables. Set brgemm output buffer.
458a1240
fix arm testcase fail issue: Fix ARM testcase failure: Retain the C++…
dfdcdbbf
Add test case compare with reference implementation
2939bc6d
Extend pagedAttention functional test to support xattention. Remove c…
e0832a6a
Minor change in test case and add condition check
4cb9e882
Add transpose.hpp
e30f1cc4
Apply review comments
defaa5f6
Fix CI failure
8c005254
maxnick
approved these changes
on 2025-11-27
Assignees
Labels
category: CPU
category: build
Milestone
Login to write a write a comment.
Login via GitHub