[MOD-13325] Add query preprocessing to QuantPreprocessor for asymmetric distance computation
#876
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
meiravgri
changed the title
meiravgri
changed the title
QuantPreprocessor for asymmetric distance computation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #876 +/- ##
==========================================
- Coverage 97.06% 97.05% -0.01%
==========================================
Files 126 126
Lines 7458 7512 +54
==========================================
+ Hits 7239 7291 +52
- Misses 219 221 +2 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
dor-forer
reviewed
Jan 4, 2026
| } | ||
| } | ||
|
|
||
| DataType sum_fast(const DataType *p) const { |
Collaborator
There was a problem hiding this comment.
Can we use it in the quantize function in some way? Or quantize doesn't need fast calculations like the query.
Collaborator
Author
There was a problem hiding this comment.
we always need faster calculations!
good catch
Co-authored-by: dor-forer <[email protected]>
dor-forer
approved these changes
Jan 4, 2026
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Outdated suggestions cannot be applied.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.
Motivation
In asymmetric scalar quantization, the storage vectors are quantized to
uint8_twhile query vectors remain as floats. The distance formulas require precomputed sums from the query vector:IP(x, y) = min * Σy_i + delta * Σ(q_i * y_i)— requiresy_sum = Σy_i||x - y||² = x_sum_squares - 2 * IP(x, y) + y_sum_squares— requiresy_sum_squares = Σy_i²Previously, these values would need to be computed during each distance calculation. By precomputing them during query preprocessing, we move this cost to a one-time operation per query.
New Functionality
QuantPreprocessor::preprocessQuerynow:(dim + 1) * sizeof(DataType)y_sum = Σy_iy_sum_squares = Σy_i²Query blob layout:
Usage in Distance Calculator
The asymmetric distance function can now retrieve the precomputed value directly from the query blob at offset
dim, avoiding redundant summation during each distance computation.