Encapsulate quantized and full-precision vector access in composite FloatVectorValues#3271
Draft
VijayanB wants to merge 1 commit into
Draft
Encapsulate quantized and full-precision vector access in composite FloatVectorValues#3271VijayanB wants to merge 1 commit into
VijayanB wants to merge 1 commit into
Conversation
…FloatVectorValues
Introduce QuantizedFloatVectorValues as an inner class of NativeEngines990KnnVectorsReader,
following the same pattern as Lucene's ScalarQuantizedVectorValues. This composite bundles
full-precision float vectors with quantized byte vectors into a single FloatVectorValues:
- vectorValue(ord) returns full-precision floats for merge/flush
- scorer(float[]) quantizes the query and delegates to quantized byte values for search
- rescorer(float[]) delegates to raw float values for full-fidelity rescoring
For ADC fields, scorer(float[]) delegates directly to FaissScorableByteVectorValues which
handles asymmetric float-vs-byte scoring via its FlatVectorsScorer. For non-ADC (Hamming/BQ),
the float query is quantized to bytes using the cached QuantizationState before scoring.
Key changes:
- NativeEngines990KnnVectorsReader.getFloatVectorValues() returns the composite when
quantization is configured, plain float values otherwise
- Extract getQuantizationState() helper to deduplicate state loading across search() and
the composite's scorer()
- Add scorer(float[]) to FaissScorableByteVectorValues for ADC asymmetric scoring
- Simplify ExactSearcher by removing manual quantization/ADC transform logic, delegating
to the composite's scorer instead
- Remove shouldRetrieveQuantizedVectors bypass in KNNVectorValuesFactory
Signed-off-by: Vijayan Balasubramanian <balasvij@amazon.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
1 participant
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.
Description
Introduce QuantizedFloatVectorValues as an inner class of NativeEngines990KnnVectorsReader,
following the same pattern as Lucene's ScalarQuantizedVectorValues. This composite bundles
full-precision float vectors with quantized byte vectors into a single FloatVectorValues:
For ADC fields, scorer(float[]) delegates directly to FaissScorableByteVectorValues which
handles asymmetric float-vs-byte scoring via its FlatVectorsScorer. For non-ADC (Hamming/BQ),
the float query is quantized to bytes using the cached QuantizationState before scoring.
Key changes:
Related Issues
Resolves #[Issue number to be closed when this PR is merged]
Check List
--signoff.By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.