[MOD-9557] Fix incorrect vector blob size calculation #665

meiravgri · 2025-04-27T11:27:11Z

Description:

This PR fixes a bug where vector data was copied using an incorrect size, leading to truncated vectors and corrupted distance calculations in certain configurations.

Bug Summary:

Previously, when copying a vector blob, we computed the size manually using:

memcpy(blob_copy.get(), this->frontendIndex->getDataByInternalId(job->id),
       this->frontendIndex->getDim() * sizeof(DataType));

This approach ignores additional data stored alongside the vector. In particular, when using int8 with cosine distance, a float-typed norm (4 bytes) is appended to the end of each vector. By not including this in the copy, the norm was lost.

This led to:

Invalid norm values (0 or garbage) during distance computation.
-inf similarity scores due to division by invalid norms.
Many vectors selecting the same neighbors.
Lock contention on shared nodes in multithreaded environments, significantly impacting performance.

Fix:

We now use the correct size returned by getDataSize() for the copy:

auto blob_copy = this->getAllocator()->allocate_unique(this->frontendIndex->getDataSize());

memcpy(blob_copy.get(), this->frontendIndex->getDataByInternalId(job->id),
       this->frontendIndex->getDataSize());

Miscellaneous Improvements

Eliminated unnecessary memory allocation and copying in `TieredHNSW_BatchIterator`

Previously, TieredHNSW_BatchIterator allocated memory and copied the original query blob, but never used this copy directly.
The internal batch iterators (flat_iterator and hnsw_iterator) create their own copies:
- flat_iterator copy is created during TieredHNSW_BatchIterator construction
- When TieredHNSW_BatchIterator::getNextResults() is called and hnsw_iterator is not initialized, it takes the blob from flat_iterator

Test utilities

Added getStoredVectorDataByLabel Method: Introduced a new method in BruteForceIndex, HNSWIndex, and their derived classes. This new method retrieves the full stored vector, including appended metadata (e.g., norms).
This complements the existing getDataByLabel method, which retrieves only the vector elements.

📝 Note:

This fix also reveals a bug in vector preprocessing where the code assumes the input blob and processed blob have the same size:

storage_blob = this->allocator->allocate(processed_bytes_count);
memcpy(storage_blob, original_blob, processed_bytes_count);

This is problematic because:

For int8 vectors with cosine similarity, processed_bytes_count includes space for the norm
But original_blob might only contain the vector elements (dim * sizeof(DataType))
This may result in out-of-bounds memcpy reads.

This issue will be fully addressed in a follow-up PR, see MOD-9576.

In this PR, preliminary steps have been taken:

TODO added: A comment has been placed in the preprocessor logic to flag the need for handling size mismatches:
// TODO: handle original_blob_size != processed_bytes_count
🔄 Batch Iterator Modification:

To eliminate duplicated, potentially unsafe copy logic and ensure consistent handling of blob size and memory allocation, the responsibility for copying the query blob has been delegated from the batch iterator to the preprocessor.

Previously, the batch iterator manually allocated memory and copied the input before invoking preprocessQueryInPlace():

auto *queryBlobCopy =
    this->allocator->allocate_aligned(this->dataSize, this->preprocessors->getAlignment());
memcpy(queryBlobCopy, queryBlob, this->dim * sizeof(DataType));
this->preprocessQueryInPlace(queryBlobCopy);

This has now been replaced with a call to the existing preprocessQuery() method with force_copy = true:

auto queryBlobCopy = this->preprocessQuery(queryBlob, true);
auto *queryBlobCopyPtr = queryBlobCopy.release();  // Take ownership

size_t main_vector_indexing_count;

format comment out new tests

codecov · 2025-04-28T14:39:36Z

Codecov Report

Attention: Patch coverage is 91.25000% with 7 lines in your changes missing coverage. Please review.

Project coverage is 96.24%. Comparing base (b22e5c2) to head (1747911).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/VecSim/algorithms/svs/svs.h	60.00%	4 Missing ⚠️
src/VecSim/algorithms/hnsw/hnsw_multi.h	85.71%	2 Missing ⚠️
.../VecSim/algorithms/brute_force/brute_force_multi.h	90.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #665      +/-   ##
==========================================
- Coverage   96.34%   96.24%   -0.11%     
==========================================
  Files         112      112              
  Lines        6239     6278      +39     
==========================================
+ Hits         6011     6042      +31     
- Misses        228      236       +8

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

increases seed every call so we will get different vectors. this required a change in metrics_test because now the score is not exactly 0. VecSimIndexAbstract CastToBruteForce()-> rename to GetFlatBufferIndex add BruteForceIndex CastToBruteForce() add getFlatBufferIndexAsBruteForce to tiered index to get BruteForceIndex functions added to hnsw and BF: Formalize getDataByLabel definition: populate the intput empty vector with the vectors data,not including any additional meta data that might present in the vector storage like the norm. add getStoredVectorDataByLabel returns a vector of char vectors. each vector contains the vector as it is stored in the vectors container, including the meta data if exists.

…nt8_anamaly

assert in FP32SpacesOptimizationTest::FP32InnerProductTest

use preprocessQuery with force_copy in batch iterator instead on preprocessQueryInPlace

// TODO: handle original_blob_size != processed_bytes_count

Copilot

Pull Request Overview

This PR addresses several improvements in testing, data processing, and memory management for vector similarity indexes. Key changes include:

Updates to test cases in both INT8 and BruteForce tests to ensure const correctness and more precise floating‑point comparisons.
Modifications to index and preprocessor APIs to allow forced data copying during query preprocessing, as well as the addition of helper functions for retrieving complete stored vector data.
Minor documentation and spelling corrections across multiple modules.

Reviewed Changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/unit/test_int8.cpp	Adjusted random vector generation and improved float assertions.
tests/unit/test_bruteforce.cpp	Introduced const correctness when fetching internal vector data.
src/VecSim/vec_sim_tiered_index.h	Added new helper to access the brute-force view of the flat buffer.
src/VecSim/vec_sim_index.h	Updated preprocessQuery signature to support forced copying.
src/VecSim/spaces/computer/preprocessors.h	Added TODO comments regarding blob size mismatches.
src/VecSim/spaces/computer/preprocessor_container.h	Updated method signatures to support a force_copy option.
src/VecSim/spaces/computer/preprocessor_container.cpp	Updated memory copy logic with force_copy and added TODO comments.
src/VecSim/algorithms/svs/svs.h	Modified query blob handling in newBatchIterator for forced copying.
src/VecSim/algorithms/hnsw/*.h	Improved vector data copying to exclude metadata and fixed spelling.
src/VecSim/algorithms/brute_force/*.h	Updated tests to return complete vector data with const correctness.
.github/workflows/event-pull_request.yml	Minor updates to build flags and commented out an unused workflow dependency.

Comments suppressed due to low confidence (1)

tests/unit/test_int8.cpp:396

[nitpick] Verify that the tolerance value (1e-6) used in ASSERT_NEAR is appropriate for the expected precision of floating‑point computations in this test scenario.

        ASSERT_NEAR(score, expected_score, 1e-6f) << "failed at vector id:" << id;

Copilot · 2025-05-04T13:14:02Z

src/VecSim/spaces/computer/preprocessors.h

@@ -55,9 +55,11 @@ class CosinePreprocessor : public PreprocessorInterface {
            // If one of them is null, allocate memory for it and copy the original_blob to it.
            if (storage_blob == nullptr) {
                storage_blob = this->allocator->allocate(processed_bytes_count);
+                // TODO: handle original_blob_size != processed_bytes_count


[nitpick] Consider elaborating on how to handle cases where the original blob size differs from processed_bytes_count to improve future maintainability.

Copilot · 2025-05-04T13:14:03Z

src/VecSim/algorithms/svs/svs.h

-            this->allocator->allocate_aligned(this->dataSize, this->preprocessors->getAlignment());
-        memcpy(queryBlobCopy, queryBlob, this->dim * sizeof(DataType));
-        this->preprocessQueryInPlace(queryBlobCopy);
+        // force_copy == true.


[nitpick] Clarify in documentation the rationale for forcing a copy (force_copy == true) of the query blob in the newBatchIterator to make the memory ownership semantics explicit.

Suggested change

// force_copy == true.

// force_copy == true. Forcing a copy ensures that the VecSimBatchIterator takes ownership

// of the query blob's memory and is responsible for freeing it. This avoids potential

// memory management issues and makes ownership semantics explicit.

…nt8_anamaly

GuyAv46

Nice catch and nice tests!

src/VecSim/algorithms/brute_force/brute_force_multi.h

GuyAv46 · 2025-05-04T16:16:35Z

src/VecSim/algorithms/brute_force/brute_force.h

+     * @return A vector containing the complete vector data (elements + metadata) for the given
+     * label
+     */
+    virtual std::vector<std::vector<char>> getStoredVectorDataByLabel(labelType label) const = 0;


Can we move the two helpers under VecSimIndexAbstract (test only), so we don't need to redefine it in HNSW?

Consider renaming - I'd expect getDataByLabel to return all the data. Maybe getDataByLabel and getVectorByLabel, or something similar

Agreed — getDataByInternalId should also be unified, as it's currently declared separately in each algorithm with different return types for BruteForce and HNSW.

I also agree that getDataByLabel isn’t a very intuitive name. getStoredVectorDataByLabel is more descriptive for the full blob, including metadata like norms.

That said, to avoid overloading in this PR, I suggest we address the renaming and unification in a separate follow-up. That would also be a good opportunity to review existing usages of getDataByLabel and see where getStoredVectorDataByLabel would be more appropriate.

GuyAv46 · 2025-05-04T16:48:21Z

src/VecSim/spaces/computer/preprocessor_container.cpp

+
+    if (needs_copy) {
+        auto aligned_mem = this->allocator->allocate_aligned(blob_bytes_count, this->alignment);
+        // TODO: handle original_blob_size != processed_bytes_count


When? Is there something that will fail a PR if we forget to support it?

ASAP. Reading beyond an allocated memory is an undefined behaviour that potentially can lead to a crash

src/VecSim/spaces/computer/preprocessors.h

…ract

…nt8_anamaly

alonre24

Nice!

alonre24 · 2025-05-05T10:20:25Z

src/VecSim/vec_sim_tiered_index.h

@@ -110,6 +110,9 @@ class VecSimTieredIndex : public VecSimIndexInterface {
    inline VecSimIndexAbstract<DataType, DistType> *getFlatBufferIndex() {
        return this->frontendIndex;
    }
+    inline BruteForceIndex<DataType, DistType> *getFlatBufferIndexAsBruteForce() {


Consider changing getFlatBufferIndex to return BruteForceIndex insread

makes sense & done

alonre24 · 2025-05-05T10:56:05Z

src/VecSim/algorithms/hnsw/hnsw_tiered.h

    VecSimQueryParams *queryParams, std::shared_ptr<VecSimAllocator> allocator)
-    : VecSimBatchIterator(query_vector, queryParams ? queryParams->timeoutCtx : nullptr,
+    // Tiered batch iterator doesn't hold its own copy of the query vector.


Why is that? Please extend the document here since it is not trivial..

src/VecSim/vec_sim_index.h

currently testing get_stored_vector_data_single_test

Copilot

Pull Request Overview

This PR fixes an issue with vector blob copying by switching from a size computed as (dim * sizeof(DataType)) to the correct size provided by getDataSize(), thereby ensuring that additional metadata (e.g. norms for int8 vectors with cosine similarity) is correctly copied. It also removes redundant memory allocation in the batch iterator and introduces new test utilities for retrieving the full stored vector data.

Reviewed Changes

Copilot reviewed 18 out of 19 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/utils/tests_utils.h	Adds a helper to populate float vectors for testing.
tests/unit/test_int8.cpp	Updates random vector population and adjusts float comparison in search tests.
tests/unit/test_index_test_utils.cpp	Improves test utilities for vector data verification.
tests/unit/test_bruteforce.cpp	Applies const-correctness to getDataByInternalId calls.
src/VecSim/vec_sim_tiered_index.h	Updates getFlatBufferIndex to return a BruteForceIndex pointer.
src/VecSim/vec_sim_index.h	Refactors preprocessQuery to incorporate a force_copy parameter.
src/VecSim/spaces/computer/*	Propagates force_copy changes and adds TODOs regarding blob size mismatches.
src/VecSim/algorithms/*	Updates vector copy logic to copy only the vector elements where appropriate and adds methods for retrieving full stored data.

Files not reviewed (1)

tests/unit/CMakeLists.txt: Language not supported

Copilot · 2025-05-06T12:53:51Z

src/VecSim/algorithms/hnsw/hnsw_tiered.h

-           this->frontendIndex->getDim() * sizeof(DataType));
+    size_t data_size = this->frontendIndex->getDataSize();
+    auto blob_copy = this->getAllocator()->allocate_unique(data_size);
+    // Assuming the size of the blob stored in the frontend index matches the size of the blob


[nitpick] Consider adding an assertion or runtime check to verify that the blob size from the frontend index (getDataSize()) matches the expected size in the HNSW index. This extra check would help catch unexpected mismatches early.

Copilot · 2025-05-06T12:53:51Z

src/VecSim/spaces/computer/preprocessors.h

+                // TODO: handle original_blob_size != processed_bytes_count
                memcpy(storage_blob, original_blob, processed_bytes_count);
            } else if (query_blob == nullptr) {
                query_blob = this->allocator->allocate_aligned(processed_bytes_count, alignment);
+                // TODO: handle original_blob_size != processed_bytes_count


Consider adding a check or assertion to ensure that the original blob size is at least as large as processed_bytes_count before calling memcpy. This can help prevent potential out-of-bounds reads if the assumption is violated.

Suggested change

// TODO: handle original_blob_size != processed_bytes_count

memcpy(storage_blob, original_blob, processed_bytes_count);

} else if (query_blob == nullptr) {

query_blob = this->allocator->allocate_aligned(processed_bytes_count, alignment);

// TODO: handle original_blob_size != processed_bytes_count

assert(original_blob_size >= processed_bytes_count && "original_blob is too small");

memcpy(storage_blob, original_blob, processed_bytes_count);

} else if (query_blob == nullptr) {

query_blob = this->allocator->allocate_aligned(processed_bytes_count, alignment);

assert(original_blob_size >= processed_bytes_count && "original_blob is too small");

Copilot · 2025-05-06T12:53:51Z

src/VecSim/algorithms/svs/svs.h

-    virtual void fitMemory() {};
+    void fitMemory() override {}
+    std::vector<std::vector<char>> getStoredVectorDataByLabel(labelType label) const override {
+        assert(nullptr && "Not implemented");


[nitpick] If getStoredVectorDataByLabel in SVSIndex is likely to be invoked during tests or in production, consider providing a safe default implementation instead of an assert, or clearly document that it should not be used.

Suggested change

assert(nullptr && "Not implemented");

// This method is not implemented. Returning an empty vector as a safe default.

…nt8_anamaly

GuyAv46

Nicely done!
Lets not forget the renaming of the getData helpers

github-actions · 2025-05-07T06:55:54Z

Backport failed for 8.0, because it was unable to cherry-pick the commit(s).

Please cherry-pick the changes locally and resolve any conflicts.

git fetch origin 8.0
git worktree add -d .worktree/backport-665-to-8.0 origin/8.0
cd .worktree/backport-665-to-8.0
git switch --create backport-665-to-8.0
git cherry-pick -x 98cf0a1174a6ffdd91e864aa6e29a11be90baf6d

* add size_t bg_vector_indexing_count; size_t main_vector_indexing_count; * fix datasize in executeInsertJob * disable neighbours print * replace manual blob size calcaulations format comment out new tests * INT8Test::PopulateRandomVector increases seed every call so we will get different vectors. this required a change in metrics_test because now the score is not exactly 0. VecSimIndexAbstract CastToBruteForce()-> rename to GetFlatBufferIndex add BruteForceIndex CastToBruteForce() add getFlatBufferIndexAsBruteForce to tiered index to get BruteForceIndex functions added to hnsw and BF: Formalize getDataByLabel definition: populate the intput empty vector with the vectors data,not including any additional meta data that might present in the vector storage like the norm. add getStoredVectorDataByLabel returns a vector of char vectors. each vector contains the vector as it is stored in the vectors container, including the meta data if exists. * build test with VERBOSE=1 * verbose cov * verbose cov fix assert in FP32SpacesOptimizationTest::FP32InnerProductTest * fix * add force_copy to maybeCopyToAlignedMem and preprocessQuery use preprocessQuery with force_copy in batch iterator instead on preprocessQueryInPlace * also batch in svs * revert redundancies * tiered batch iterator doesn't copy the blob * add // TODO: handle original_blob_size != processed_bytes_count * revert pull request changes * move getDataByLabel and getStoredVectorDataByLabel to VecSimIndexAbstract * add test_index_test_utils currently testing get_stored_vector_data_single_test * getFlatBufferIndex returns BruteForceIndex * little test name fix * small fix * fix range (cherry picked from commit 98cf0a1)

[MOD-9557] Fix incorrect vector blob size calculation (#665) * add size_t bg_vector_indexing_count; size_t main_vector_indexing_count; * fix datasize in executeInsertJob * disable neighbours print * replace manual blob size calcaulations format comment out new tests * INT8Test::PopulateRandomVector increases seed every call so we will get different vectors. this required a change in metrics_test because now the score is not exactly 0. VecSimIndexAbstract CastToBruteForce()-> rename to GetFlatBufferIndex add BruteForceIndex CastToBruteForce() add getFlatBufferIndexAsBruteForce to tiered index to get BruteForceIndex functions added to hnsw and BF: Formalize getDataByLabel definition: populate the intput empty vector with the vectors data,not including any additional meta data that might present in the vector storage like the norm. add getStoredVectorDataByLabel returns a vector of char vectors. each vector contains the vector as it is stored in the vectors container, including the meta data if exists. * build test with VERBOSE=1 * verbose cov * verbose cov fix assert in FP32SpacesOptimizationTest::FP32InnerProductTest * fix * add force_copy to maybeCopyToAlignedMem and preprocessQuery use preprocessQuery with force_copy in batch iterator instead on preprocessQueryInPlace * also batch in svs * revert redundancies * tiered batch iterator doesn't copy the blob * add // TODO: handle original_blob_size != processed_bytes_count * revert pull request changes * move getDataByLabel and getStoredVectorDataByLabel to VecSimIndexAbstract * add test_index_test_utils currently testing get_stored_vector_data_single_test * getFlatBufferIndex returns BruteForceIndex * little test name fix * small fix * fix range (cherry picked from commit 98cf0a1)

meiravgri added 4 commits April 15, 2025 11:55

add size_t bg_vector_indexing_count;

25f3c3a

size_t main_vector_indexing_count;

fix datasize in executeInsertJob

ffe9e81

disable neighbours print

64b3d3f

replace manual blob size calcaulations

94e5789

format comment out new tests

meiravgri added 11 commits April 30, 2025 04:24

Merge remote-tracking branch 'origin/main' into meiravg_investigate_i…

66a7237

…nt8_anamaly

build test with VERBOSE=1

80d5cec

verbose cov

3fe0515

verbose cov fix

b097a20

assert in FP32SpacesOptimizationTest::FP32InnerProductTest

fix

b10dd24

add force_copy to maybeCopyToAlignedMem and preprocessQuery

6b2a89e

use preprocessQuery with force_copy in batch iterator instead on preprocessQueryInPlace

also batch in svs

c71876b

revert redundancies

07f9ec1

tiered batch iterator doesn't copy the blob

a618046

add

a038e1d

// TODO: handle original_blob_size != processed_bytes_count

meiravgri requested a review from Copilot May 4, 2025 13:13

Copilot AI reviewed May 4, 2025

View reviewed changes

meiravgri added 2 commits May 4, 2025 13:26

revert pull request changes

3662895

Merge remote-tracking branch 'origin/main' into meiravg_investigate_i…

284646e

…nt8_anamaly

meiravgri changed the title ~~Meiravg_investigate_int8_anamaly~~ [MOD-9557] Fix blob allocations size May 4, 2025

meiravgri changed the title ~~[MOD-9557] Fix blob allocations size~~ [MOD-9557] Fix incorrect vector blob size calculation May 4, 2025

meiravgri requested review from GuyAv46 and alonre24 May 4, 2025 14:27

GuyAv46 reviewed May 4, 2025

View reviewed changes

meiravgri added 2 commits May 5, 2025 04:51

move getDataByLabel and getStoredVectorDataByLabel to VecSimIndexAbst…

5a672ea

…ract

Merge remote-tracking branch 'origin/main' into meiravg_investigate_i…

2d4e7ca

…nt8_anamaly

alonre24 reviewed May 5, 2025

View reviewed changes

meiravgri added 2 commits May 6, 2025 12:42

add test_index_test_utils

4fa002a

currently testing get_stored_vector_data_single_test

getFlatBufferIndex returns BruteForceIndex

0c3d47d

meiravgri requested review from alonre24, GuyAv46 and Copilot May 6, 2025 12:52

Copilot AI reviewed May 6, 2025

View reviewed changes

meiravgri added 3 commits May 6, 2025 12:56

little test name fix

2495ae6

small fix

1cf9456

Merge remote-tracking branch 'origin/main' into meiravg_investigate_i…

b4c0f3f

…nt8_anamaly

GuyAv46 previously approved these changes May 6, 2025

View reviewed changes

alonre24 previously approved these changes May 6, 2025

View reviewed changes

meiravgri added the backport 8.0 label May 6, 2025

meiravgri added this pull request to the merge queue May 6, 2025

github-merge-queue bot removed this pull request from the merge queue due to failed status checks May 6, 2025

fix range

1747911

meiravgri dismissed stale reviews from alonre24 and GuyAv46 via 1747911 May 7, 2025 04:42

meiravgri requested review from GuyAv46 and alonre24 May 7, 2025 04:44

meiravgri enabled auto-merge May 7, 2025 04:45

GuyAv46 approved these changes May 7, 2025

View reviewed changes

meiravgri added this pull request to the merge queue May 7, 2025

Merged via the queue into main with commit 98cf0a1 May 7, 2025
15 of 17 checks passed

meiravgri deleted the meiravg_investigate_int8_anamaly branch May 7, 2025 06:55

meiravgri mentioned this pull request May 11, 2025

[8.0] [MOD-9557] Fix incorrect vector blob size calculation #672

Merged

-        // force_copy == true.
+        // force_copy == true. Forcing a copy ensures that the VecSimBatchIterator takes ownership
+        // of the query blob's memory and is responsible for freeing it. This avoids potential
+        // memory management issues and makes ownership semantics explicit.

	assert(nullptr && "Not implemented");
	// This method is not implemented. Returning an empty vector as a safe default.

[MOD-9557] Fix incorrect vector blob size calculation #665

[MOD-9557] Fix incorrect vector blob size calculation #665

Uh oh!

Conversation

meiravgri commented Apr 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description:

Bug Summary:

Fix:

Miscellaneous Improvements

Eliminated unnecessary memory allocation and copying in TieredHNSW_BatchIterator

Test utilities

📝 Note:

This issue will be fully addressed in a follow-up PR, see MOD-9576.

Uh oh!

codecov bot commented Apr 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI May 4, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 4, 2025

Choose a reason for hiding this comment

Uh oh!

GuyAv46 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

GuyAv46 May 4, 2025

Choose a reason for hiding this comment

Uh oh!

meiravgri May 5, 2025

Choose a reason for hiding this comment

Uh oh!

meiravgri May 5, 2025

Choose a reason for hiding this comment

Uh oh!

GuyAv46 May 4, 2025

Choose a reason for hiding this comment

Uh oh!

meiravgri May 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

alonre24 left a comment

Choose a reason for hiding this comment

Uh oh!

alonre24 May 5, 2025

Choose a reason for hiding this comment

Uh oh!

meiravgri May 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alonre24 May 5, 2025

Choose a reason for hiding this comment

Uh oh!

meiravgri May 6, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI May 6, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 6, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 6, 2025

meiravgri commented Apr 27, 2025 •

edited

Loading

Eliminated unnecessary memory allocation and copying in `TieredHNSW_BatchIterator`

codecov bot commented Apr 28, 2025 •

edited

Loading

meiravgri May 4, 2025 •

edited

Loading

meiravgri May 6, 2025 •

edited

Loading