Add Unified Attention docs #275
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
madamczyk-intel
requested review from
adobrzyn,
afierka-intel,
kzawora-intel,
mgawarkiewicz-intel,
mswiniarsk,
vivekgoe and
xuechendi
as code owners
Signed-off-by: Michal Adamczyk <[email protected]>
madamczyk-intel
force-pushed
the
unified_attn_docs
branch
from
fbb3af8 to
fba9d57
Compare
adobrzyn
approved these changes
Nov 4, 2025
| The problem here lies in the denominator as it contains the sum of all terms. Fortunately we can split the calculation into two separate softmax and then readjust the results and combine them. Let's say we have: | ||
| $$z_1, z_2\text{ - local softmax results} \\ c_1, c_2 \text{ - local maxima} \\ s_1, s_2 \text{ - local sums}$$ | ||
| We can then calculate: | ||
| $$c = max(c_1, c_2) \\ adj_i = e^{c_i-c} \\ s = s_1 * adj_1 + s_2 * adj_2\\ z_i\prime = \frac{z_i*s_i*adj_i}{s} $$ |
Collaborator
There was a problem hiding this comment.
Precommit fails here
PatrykWo
added a commit
that referenced
this pull request
Nov 4, 2025
Signed-off-by: PatrykWo <[email protected]>
Merged
PatrykWo
added a commit
that referenced
this pull request
Nov 4, 2025
plus fix for PR #275 --------- Signed-off-by: PatrykWo <[email protected]>
PatrykWo
pushed a commit
that referenced
this pull request
Nov 5, 2025
Signed-off-by: Michal Adamczyk <[email protected]> Co-authored-by: Agata Dobrzyniewicz <[email protected]>
PatrykWo
added a commit
that referenced
this pull request
Nov 5, 2025
plus fix for PR #275 --------- Signed-off-by: PatrykWo <[email protected]>
PatrykWo
added a commit
that referenced
this pull request
Nov 7, 2025
plus fix for PR #275 --------- Signed-off-by: PatrykWo <[email protected]>
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.
No description provided.