[NPUW]Transpose V tensor for Softmax - Slice - Matmul.#33807
Merged
intelgaoxiong merged 2 commits intoopenvinotoolkit:masterfrom Jan 28, 2026
Merged
[NPUW]Transpose V tensor for Softmax - Slice - Matmul.#33807intelgaoxiong merged 2 commits intoopenvinotoolkit:masterfrom
intelgaoxiong merged 2 commits intoopenvinotoolkit:masterfrom
Conversation
github-actions
bot
added
category: NPU
Contributor
Author
rkazants
reviewed
Jan 26, 2026
|
|
||
| // llama2 pattern for value tensor concate | ||
| class TransposeValueTensors_llama2 : public TransposeValueTensors { | ||
| // MHA (Multi-Head Attention) pattern for value tensor concatenation |
Collaborator
There was a problem hiding this comment.
Can you implement test(s) for transformation?
intelgaoxiong
force-pushed
the
xiong/transpose_v_for_gpt_oss
branch
from
55c4fd5 to
68646a4
Compare
esmirno
approved these changes
Jan 26, 2026
Contributor
esmirno
left a comment
esmirno
left a comment
There was a problem hiding this comment.
thanks, unit tests are maintained in expected way
intelgaoxiong
force-pushed
the
xiong/transpose_v_for_gpt_oss
branch
from
68646a4 to
3420d33
Compare
Contributor
Author
|
@dmatveev Could you please take a look? |
Signed-off-by: intelgaoxiong <xiong.gao@intel.com>
Signed-off-by: intelgaoxiong <xiong.gao@intel.com>
Contributor
|
@intelgaoxiong in your report, the branch is shown as |
Contributor
Author
The package name is below, commit ID exactly matches the ID in this PR. The validated packaged only included the changes in this PR. |
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
Contributor
Author
PR was removed from merge queue due to CI failure in above.
Add to merge queue again to see if the failure will disappear. |
Naseer-010
pushed a commit
to Naseer-010/openvino
that referenced
this pull request
Feb 18, 2026
…it#33807) ### Details: GPT-OSS SDPA has sink input. There is a pair of Concat and Slice around Softmax. V tensor transpose could not work for the pattern. This PR extended V tensor transpose for GPT-OSS pattern to eliminate the Permutation in compiler. ### Tickets: - *[EISW-200448](https://jira.devtools.intel.com/browse/EISW-200448)* --------- Signed-off-by: intelgaoxiong <xiong.gao@intel.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Details:
GPT-OSS SDPA has sink input.
There is a pair of Concat and Slice around Softmax.
V tensor transpose could not work for the pattern.
This PR extended V tensor transpose for GPT-OSS pattern to eliminate the Permutation in compiler.
Tickets: