feat(ivf): Add numpy array support for IVF assemble functions#287
Merged
feat(ivf): Add numpy array support for IVF assemble functions#287
Conversation
- Add assemble_from_clustering and assemble_from_file methods that accept numpy arrays directly (py_data parameter) in addition to file paths - Refactor assemble_dynamic_from_clustering to avoid double data copy: - Internal _impl function takes data by const reference - Smart dispatching detects ImmutableMemoryDataset and passes by reference - Add timing instrumentation for assemble operations - Add comprehensive tests for numpy array assembly in both IVF and DynamicIVF - Zero-copy path: numpy views passed directly without data duplication This enables users to build IVF indices from in-memory numpy arrays without intermediate file I/O, improving performance for dynamic workflows.
ethanglaser
approved these changes
Mar 17, 2026
|
|
||
| /// Index building | ||
| // Build from Numpy array. | ||
| detail::add_build_specialization<svs::BFloat16>(clustering); |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR extends the IVF and DynamicIVF Python APIs to assemble indices directly from in-memory NumPy arrays (via a py_data parameter) in addition to file-based loaders, and refactors DynamicIVF assembly internals to reduce unnecessary data moves/copies while adding assembly timing instrumentation.
Changes:
- Add NumPy-array overloads for
IVF.assemble_from_clustering/IVF.assemble_from_fileand the DynamicIVF equivalents in the pybind layer. - Refactor
assemble_dynamic_from_clusteringto route already-loaded datasets by reference via an internal_implfunction and add timing/debug logging. - Add Python tests covering NumPy-based assembly for IVF and DynamicIVF.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| include/svs/orchestrators/dynamic_ivf.h | Switch assembly APIs to perfect-forward data_proto to enable non-copying paths for already-loaded datasets/views. |
| include/svs/index/ivf/dynamic_ivf.h | Add assemble_dynamic_from_clustering_impl (const-ref data) + dispatch wrapper; add timing instrumentation. |
| bindings/python/src/ivf.cpp | Add pybind overloads for assembling IVF from NumPy arrays; remove extra copy in clustering build-from-array path. |
| bindings/python/src/dynamic_ivf.cpp | Add pybind overloads for assembling DynamicIVF from NumPy arrays (with ids). |
| bindings/python/tests/test_ivf.py | Add test coverage for IVF NumPy-array assembly paths. |
| bindings/python/tests/test_dynamic_ivf.py | Add test coverage for DynamicIVF NumPy-array assembly paths (including add/delete after assembly). |
Comments suppressed due to low confidence (1)
include/svs/index/ivf/dynamic_ivf.h:1066
clustering.centroids()is called on an lvalueclustering, which will copy the centroid dataset (potentially large). Sinceclusteringis a by-value parameter and not used after buildingdense_clusters, move it when extracting centroids (e.g., use the rvalue overload) to avoid the extra copy during assembly.
Distance,
decltype(threadpool)>(
clustering.centroids(),
std::move(dense_clusters),
include/svs/index/ivf/dynamic_ivf.h
Outdated
Comment on lines
+1006
to
+1024
| const size_t intra_query_thread_count = 1 | ||
| Pool& threadpool, | ||
| const size_t intra_query_thread_count = 1, | ||
| svs::logging::logger_ptr logger = svs::logging::get() |
Comment on lines
+150
to
+155
| auto mutable_view = svs::data::SimpleDataView<T, N>( | ||
| const_cast<T*>(view.data()), view.size(), view.dimensions() | ||
| ); | ||
| return svs::IVF::assemble_from_clustering<Q>( | ||
| std::move(clustering), mutable_view, distance_type, num_threads, intra_query_threads | ||
| ); |
Comment on lines
+142
to
+147
| auto mutable_view = svs::data::SimpleDataView<T, N>( | ||
| const_cast<T*>(view.data()), view.size(), view.dimensions() | ||
| ); | ||
| return svs::DynamicIVF::assemble_from_clustering<Q>( | ||
| std::move(clustering), | ||
| mutable_view, |
include/svs/index/ivf/dynamic_ivf.h
Outdated
| @@ -1069,6 +1069,59 @@ auto assemble_dynamic_from_clustering( | |||
| std::move(threadpool), | |||
| intra_query_thread_count | |||
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
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.
This enables users to build IVF indices from in-memory numpy arrays without intermediate file I/O, improving performance for dynamic workflows.
- Internal _impl function takes data by const reference
- Smart dispatching detects ImmutableMemoryDataset and passes by reference