Skip to content

Test XAttention #32148

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Draft
wants to merge 31 commits into
base: master
Choose a base branch
from
Draft

Test XAttention #32148

wants to merge 31 commits into from
+5,552 −328

Conversation

peterchen-intel
Copy link
Contributor

Details:

  • item1
  • ...

Tickets:

  • ticket-id

@github-actions github-actions bot added the category: GPU OpenVINO GPU plugin label Sep 18, 2025
riverlijunjie and others added 27 commits September 26, 2025 14:51
@riverlijunjie @ceciliapeng2011
Init PA CM Impl(1st/2nd token and kvcache update)
e030c80
@riverlijunjie @ceciliapeng2011
enabled simple pa unit tests pass
435a7ac
@riverlijunjie @ceciliapeng2011
Fix 2nd_token issue
8947906
@riverlijunjie @ceciliapeng2011
Fixed pipeline output corruption issue
83dba29 
   1. kvcache update's k/v offset issue
   2. 2nd token lse data overflow issue
@riverlijunjie @ceciliapeng2011
Fix 2nd non-16 alignment accuracy issue
2743aab
@riverlijunjie @ceciliapeng2011
Set best partition size for 2nd
65b9cc7
@ceciliapeng2011
update KV_BLOCK_SIZE to 256
c4a1659
@ceciliapeng2011
initiate xattention integration
62a222f
@ceciliapeng2011
qwen2.5-1.5b 4k trunk works with xatten.
ac882ab
@ceciliapeng2011
4k aligned works.
0621e4b
@ceciliapeng2011
fix block_mask not fully initialized issue.
98a4ecd
@ceciliapeng2011
fix of find_block
5af3330
@ceciliapeng2011
xatten: fix accuacy problem caused by debug
4f9ed28
@luo-cheng2021 @ceciliapeng2011
use int32 to store float INV_S to align python version accuracy
d35f4fb
@ceciliapeng2011
OV_GPU_XATTN_BLOCK_SIZE and OV_GPU_XATTN_THRESH
4e25a4a
@usstq @ceciliapeng2011
fix building error on windows.
c3c87b7
@ceciliapeng2011
process tail in find_block
76685f0
@riverlijunjie @ceciliapeng2011
Fix f16 accuracy issue and optimize 2nd token to improve 5%
c5bdcf9
@ceciliapeng2011
fix waring_as_error on CI Windows.
95a2da1
@ceciliapeng2011
dump block mask with DUMP_XATTN_BLOCK_MASK for debug
36bee72
@riverlijunjie @ceciliapeng2011
Support kv cache u8 precision
4fa97be
@ceciliapeng2011
refactor: split into pa_common and sdpa_common, which include attenti…
55ba7c3 
…on_common.
@ceciliapeng2011
integrate xattn_post_proc kernel and FP16 kernel works. TODOto verify…
a06adef 
… u8 kvcache.
@riverlijunjie @ceciliapeng2011
update partition size
4b391be
@ceciliapeng2011
enable int8 kvcache for xatten, but accuracy fails.
f2f2126
@ceciliapeng2011
fix xattn kvcache u8 accuracy issue.
89c8577
@riverlijunjie @ceciliapeng2011
Fix 2nd accuracy issue
024b71a
@github-actions github-actions bot added the category: transformations OpenVINO Runtime library - Transformations label Sep 30, 2025
riverlijunjie and others added 2 commits October 10, 2025 09:20
@riverlijunjie
Optimize single_token_finalization kernel with fixed unroll
13b1122
@peterchen-intel
Fix Win build issue
2a62363 
 src/plugins/intel_gpu/tests/unit/test_cases/paged_attention_gpu_test.cpp
1253 line

Signed-off-by: Chen, Peter <[email protected]>
@peterchen-intel peterchen-intel force-pushed the test/xattention branch 2 times, most recently from 0103862 to 2a62363 Compare October 10, 2025 12:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

category: GPU OpenVINO GPU plugin category: transformations OpenVINO Runtime library - Transformations do_not_merge do_not_review

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@peterchen-intel @riverlijunjie @ceciliapeng2011 @luo-cheng2021 @usstq