Skip to content

Wrap dropped-FP scalar-quantized float values so scoring stays quantized #16415

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Pulkitg64 wants to merge 1 commit into
base: main
Choose a base branch
from
+43 −27
Open

Wrap dropped-FP scalar-quantized float values so scoring stays quantized #16415

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 21 additions & 13 deletions ...java/org/apache/lucene/backward_codecs/lucene99/Lucene99ScalarQuantizedVectorsReader.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,19 +194,6 @@ private FieldEntry getFieldEntry(String field) {
public FloatVectorValues getFloatVectorValues(String field) throws IOException {
final FieldEntry fieldEntry = getFieldEntry(field);
final FloatVectorValues rawVectorValues = rawVectorsReader.getFloatVectorValues(field);
if (rawVectorValues.size() == 0) {
return OffHeapQuantizedFloatVectorValues.load(
fieldEntry.ordToDoc,
fieldEntry.dimension,
fieldEntry.size,
fieldEntry.scalarQuantizer,
fieldEntry.similarityFunction,
vectorScorer,
fieldEntry.compress,
fieldEntry.vectorDataOffset,
fieldEntry.vectorDataLength,
quantizedVectorData);
}

OffHeapQuantizedByteVectorValues quantizedByteVectorValues =
OffHeapQuantizedByteVectorValues.load(
Expand All @@ -220,6 +207,27 @@ public FloatVectorValues getFloatVectorValues(String field) throws IOException {
fieldEntry.vectorDataOffset,
fieldEntry.vectorDataLength,
quantizedVectorData);

if (rawVectorValues.size() == 0) {
// Full-precision vectors were dropped. Wrap the dequantizing read view with the quantized
// values so scorer() stays quantized (as in the full-precision branch) while
// vectorValue()/rescorer() dequantize. Passing the dequantized view straight to a scorer
// would misroute to the non-quantized flat scorer over the quantized slice.
FloatVectorValues dequantizedRawVectorValues =
OffHeapQuantizedFloatVectorValues.load(
fieldEntry.ordToDoc,
fieldEntry.dimension,
fieldEntry.size,
fieldEntry.scalarQuantizer,
fieldEntry.similarityFunction,
vectorScorer,
fieldEntry.compress,
fieldEntry.vectorDataOffset,
fieldEntry.vectorDataLength,
quantizedVectorData);
return new QuantizedVectorValues(dequantizedRawVectorValues, quantizedByteVectorValues);
}

return new QuantizedVectorValues(rawVectorValues, quantizedByteVectorValues);
}

Expand Down
36 changes: 22 additions & 14 deletions ...re/src/java/org/apache/lucene/codecs/lucene104/Lucene104ScalarQuantizedVectorsReader.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -245,20 +245,6 @@ public FloatVectorValues getFloatVectorValues(String field) throws IOException {

FloatVectorValues rawFloatVectorValues = rawVectorsReader.getFloatVectorValues(field);

if (rawFloatVectorValues.size() == 0) {
return OffHeapScalarQuantizedFloatVectorValues.load(
fi.ordToDocDISIReaderConfiguration,
fi.dimension,
fi.size,
fi.scalarEncoding,
fi.similarityFunction,
vectorScorer,
fi.centroid,
fi.vectorDataOffset,
fi.vectorDataLength,
quantizedVectorData);
}

OffHeapScalarQuantizedVectorValues sqvv =
OffHeapScalarQuantizedVectorValues.load(
fi.ordToDocDISIReaderConfiguration,
Expand All @@ -273,6 +259,28 @@ public FloatVectorValues getFloatVectorValues(String field) throws IOException {
fi.vectorDataOffset,
fi.vectorDataLength,
quantizedVectorData);

if (rawFloatVectorValues.size() == 0) {
// Full-precision vectors were dropped. Wrap the dequantizing read view with sqvv so scorer()
// stays quantized (as in the full-precision branch) while vectorValue()/rescorer()
// dequantize.
// Passing the dequantized view straight to a scorer would misroute to the non-quantized flat
// scorer over the quantized slice.
FloatVectorValues dequantizedRawVectorValues =
OffHeapScalarQuantizedFloatVectorValues.load(
fi.ordToDocDISIReaderConfiguration,
fi.dimension,
fi.size,
fi.scalarEncoding,
fi.similarityFunction,
vectorScorer,
fi.centroid,
fi.vectorDataOffset,
fi.vectorDataLength,
quantizedVectorData);
return new ScalarQuantizedVectorValues(dequantizedRawVectorValues, sqvv);
}

return new ScalarQuantizedVectorValues(rawFloatVectorValues, sqvv);
}

Expand Down
Loading