This repository was archived by the owner on Sep 17, 2026. It is now read-only.

Optimize CAGRA_HNSW index build; add example - #195

Closed

jamxia155 wants to merge 19 commits into

NVIDIA:mainfrom

jamxia155:cagra-hnsw-build-optimization-with-example

Commits on Aug 13, 2026

cuvs-lucene__139: This code allows us to construct the HNSW graph on GPU

EC2 Default User
authored and
jamxia155
committed
47c1430
Consolidate duplicated HNSW graph and field-writing methods

nvzm123
authored and
jamxia155
committed
04df750
Add example: chunked sequential ingestion of large .fbin files
jamxia155
committed
d5b6aa7
Parallelize level-0 HNSW graph serialization in writeGraph
jamxia155
committed
288f854
Add native flat buffering for single-segment CAGRA_HNSW builds
jamxia155
committed
198e682
Parallelize CAGRA-to-HNSW conversion in GPUBuiltHnswGraph
jamxia155
committed
1976027
Add prefetching + reused-array to the fbin ingest example
jamxia155
committed
7275ed5
Honor cagraGraphBuildAlgo override in HEURISTIC strategy
jamxia155
committed
ae37031
Expand the fbin ingest example into a full optimized CAGRA-HNSW build
jamxia155
committed
5b96005
Expand OptimizedCagraHnswBuildExample and add early index-sort check
jamxia155
committed
bc3e921
Fix merge errors
jamxia155
committed
3bdeab9
Add a Lucene-version tripwire and round-trip test for NativeFlatVectorsWriter
jamxia155
committed
c942646
Add equivalence test for writerThreads-parallelized graph construction/write
jamxia155
committed
48e0ad7
Add functional coverage for native flat buffering (numInputVectors)
jamxia155
committed
5f20ffd

Commits on Aug 14, 2026