[GPU] Enable multi head size support for KV cache #29936
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
build_jenkins |
|
build_jenkins |
|
build_jenkins |
p-durandin
reviewed
Apr 7, 2025
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_base.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_micro.cpp
Outdated
Show resolved
Hide resolved
Lyamin-Roman
reviewed
Apr 7, 2025
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/pa_sdpa_kernel_opt.cpp
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_base.cpp
Outdated
Show resolved
Hide resolved
e-ddykim
reviewed
Apr 7, 2025
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_micro.cpp
Outdated
Show resolved
Hide resolved
yeonbok
reviewed
Apr 7, 2025
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/pa_kv_cache_update_ref.cl
Outdated
Show resolved
Hide resolved
yeonbok
reviewed
Apr 7, 2025
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/pa_kv_cache_update_ref.cl
Show resolved
Hide resolved
yeonbok
reviewed
Apr 7, 2025
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/sdpa_opt.cl
Outdated
Show resolved
Hide resolved
|
build_jenkins |
Lyamin-Roman
approved these changes
Apr 8, 2025
|
Found regression issue when running qwen2-7b with paged_attention. Need to check on it. |
|
build_jenkins |
e-ddykim
approved these changes
Apr 10, 2025
|
Is the regression issue resolved? If not, please add a label of "Do not merge" or "Under perf check" |
clee30
force-pushed
the
kv_multiheadsize
branch
from
2e6968e to
ec7d46f
Compare
sshlyapn
reviewed
Apr 24, 2025
...ugins/intel_gpu/tests/functional/single_layer_tests/dynamic/scaled_dot_product_attention.cpp
Show resolved
Hide resolved
src/plugins/intel_gpu/tests/unit/test_cases/paged_attention_gpu_test.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/tests/unit/test_cases/paged_attention_gpu_test.cpp
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/cl_kernels/sdpa_opt.cl
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/pa_sdpa_kernel_opt.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_opt.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_ref.cpp
Outdated
Show resolved
Hide resolved
clee30
force-pushed
the
kv_multiheadsize
branch
from
7e73669 to
3d82d72
Compare
|
build_jenkins |
clee30
force-pushed
the
kv_multiheadsize
branch
from
3d82d72 to
ba58dbe
Compare
|
build_jenkins |
clee30
force-pushed
the
kv_multiheadsize
branch
2 times, most recently
from
b9153fc to
d2a58d5
Compare
|
build_jenkins |
sshlyapn
reviewed
Apr 28, 2025
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/pa_sdpa_kernel_opt.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/pa_sdpa_kernel_opt.cpp
Outdated
Show resolved
Hide resolved
src/plugins/intel_gpu/src/kernel_selector/kernels/sdpa/sdpa_kernel_opt.cpp
Outdated
Show resolved
Hide resolved
clee30
force-pushed
the
kv_multiheadsize
branch
from
d2a58d5 to
553897f
Compare
|
build_jenkins |
clee30
force-pushed
the
kv_multiheadsize
branch
from
553897f to
e2d33bb
Compare
|
build_jenkins |
clee30
force-pushed
the
kv_multiheadsize
branch
from
a7bf0f7 to
f6d4cf7
Compare
|
build_jenkins |
sshlyapn
reviewed
May 1, 2025
...ugins/intel_gpu/tests/functional/single_layer_tests/dynamic/scaled_dot_product_attention.cpp
Outdated
Show resolved
Hide resolved
sshlyapn
approved these changes
May 1, 2025
In continue batching, head size for key and value will be different. Add support for it for sdpa and paged attention. sdpa_opt has updated to work correctly for SDPA. Besides, force multi head for dGPU to use sdpa_opt as sdpa_micro doesn't work yet
|
build_jenkins |
p-durandin
approved these changes
May 1, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In continue batching, head size for key and value will be different. Add support for it for sdpa.
Tickets:
CVS-162339 and CVS-161089