v2.2.7

What's Changed

• fix: [cherry-pick]Replace gettid() with syscall(SYS_gettid) by @xige-16 in #697

New Contributors

• @xige-16 made their first contribution in #697

Full Changelog: v2.2.6...v2.2.7

v2.3.6

What's Changed

• Add data type conversion test by @cydrain in #653
• Update BruteForce ut for multi data type by @cydrain in #656
• Clean build warnings by @cydrain in #657
• Support numpy array to sparse dataset in pyknowhere api by @chasingegg in #654
• cagra support train on gpu, search on cpu, use hnswlib by @Presburger in #661
• Add following APIs to facilitate DataSet conversion by @cydrain in #659
• Update all APIs' input parameter to 'const DataSetPtr' by @cydrain in #667
• not check diskann range_search params max_k < min_k if with_cardinal by @alwayslove2013 in #664
• fix aix compilation by @alexanderguzhva in #665
• propagate faiss changes from (24 Jun 2024) by @alexanderguzhva in #663
• [range search] iterator stops only if exceeds radius several times consecutively. by @alwayslove2013 in #671
• [skip e2e] Automated Update Cardinal Commit by @sre-ci-robot in #673
• refactor memory management by @alexanderguzhva in #569
• remove additional metric_type definition from diskann_config by @alwayslove2013 in #674
• [fix] serialize on empty index by @foxspy in #677
• [skip e2e] Automated Update Cardinal Commit by @sre-ci-robot in #680
• a theoretical fix for binary data type by @alexanderguzhva in #676
• Fix uneven loading of gpu memory when use multi device by @Presburger in #681
• [skip ci] Update cardinal version by @foxspy in #684
• fix possible memory leak caused by iterator by @zhengbuqian in #689
• fix diskann range_search config: search_list_size should for_range_search by @alwayslove2013 in #694

Full Changelog: v2.3.5...v2.3.6

v2.3.5

What's Changed

• fix: make GetIsSparse const by @zhengbuqian in #627
• check the combination of index type and data type by @cqy123456 in #634
• add sparse support type in knowherecheck by @cqy123456 in #635
• [skip ci] Add diskann benchmark test by @cydrain in #637
• register index(binary_hnsw) in knowhere by @cqy123456 in #638
• [skip e2e] : remove iterator/sparse ut limit for cardinal by @foxspy in #636
• Params for iterator by @alwayslove2013 in #625
• [skip ci] Fix diskann benchmark fail by @cydrain in #640
• update raft to 24.04 by @Presburger in #628
• [skip ci] Update range search benchmark result display by @cydrain in #646
• [skip ci] Update script prepare_gpu_build.sh for nightly build by @cydrain in #648
• Update index and data type check function by @cqy123456 in #649

Full Changelog: v2.3.4...v2.3.5

v2.3.4

What's Changed

• Dockerfile optimizations
• Add Size interface for binaryset
• Replace binary hnsw when cardinal is enabled
• Removed seed_ef for iterator and use ef directly for HNSW
• Fix iterator based range search for similarity metric when range filter is not provided

New Contributors

• @chyezh made their first contribution in #586

Full Changelog: v2.3.3...v2.3.4

v2.3.2

What's Changed

• [cherry-pick] AIX-related patches from the Faiss baseline by @alexanderguzhva in #526
• [cherrypick] Unroll loop in lookup_2_lanes for PQFasfScan for PPC64 by @alexanderguzhva in #524
• fix compiler warnings by @alexanderguzhva in #527
• sparse float vector index to support mmap when loading from index file by @zhengbuqian in #530
• Bitset valid check by @chasingegg in #532
• Update cardinal link rule for better development convenience by @foxspy in #533
• [automated] Update Cardinal Commit by @sre-ci-robot in #536
• [skip e2e] Automated Update Cardinal Commit by @sre-ci-robot in #540
• Fix LAPACK not found by @chasingegg in #542
• add a no gpu memory pool resource by @Presburger in #514
• Add binary support for pyknowhere by @foxspy in #549
• Binary vector from/to np.uint8 array in pyknowhere api by @chasingegg in #550
• fix compiler warnings by @alexanderguzhva in #538
• Update cardinal version by @chasingegg in #553
• [Bugfix] Add patch for bf16 distance computation by @foxspy in #554
• [skip e2e] Automated Update Cardinal Commit by @sre-ci-robot in #556
• Add Dockerfile for ubuntu22.04 by @cydrain in #557
• [skip ci] Update docker builder default os to ubuntu22.04 by @cydrain in #558
• only search use GPU memory pool by @Presburger in #559
• [skip e2e] Release cardinal for milvus 2.4 by @foxspy in #566

New Contributors

• @sre-ci-robot made their first contribution in #536

Full Changelog: v2.3.1...v2.3.2

v2.3.1

What's Changed

• remove two unused variables by @zhjwpku in #495
• [improvement] Make MAP_POPULATE configurable by @foxspy in #499
• Fix create index log by @chasingegg in #504
• knowhere light for milvus local by @Presburger in #502
• Fix SQ4 AVX-512 implementation by @alexanderguzhva in #506
• make IdVal more generic by @zhengbuqian in #505
• Upgrade faiss to 1.8.0 by @alexanderguzhva in #453
• Add a common subclass for IndexNode::iterator by @zhengbuqian in #501
• make DataSet gen helpers available to pyknowhere by @zhengbuqian in #509
• remove no-as-needed by @Presburger in #516
• Add range search capability to indexes with AnnIterator support by @zhengbuqian in #508
• removed unused range util for faiss by @zhengbuqian in #517
• remove accidentally added duplicate file by @zhengbuqian in #522

New Contributors

• @zhjwpku made their first contribution in #495

Full Changelog: v2.3.0...v2.3.1

v2.2.6

What's Changed

• Add bf16 distance calculator patch for zilliz-cloud (#519 ) by @foxspy in #525

Full Changelog: v2.2.5...v2.2.6

knowhere-v2.3.0

Release Notes

What's new?

• Sparse Index:
  • Introduced sparse vectors. Indexes SPARSE_WAND and SPARSE_INVERTED_INDEX are supported.
• GPU-based indexes:
  • Supported CAGRA, a GPU-based graph index from Nvidia.
  • Supported BRUTE_FORCE for GPU.
• Iterator:
  • Iterating approximately nearest vector is supported. A new API Interator is introduced for HNSW and IVF-series indexes.
  • SPARSE_WAND and SPARSE_INVERTED_INDEX also supports iterators.
• Multiple Data Types:
  • Knowhere now supports BF16 and FP16 as data type.
• Breaking Change:
  • knowhere::IndexFactory::Instance().Create<T>() now returns an expected<Index<IndexNode>> instead of Index<IndexNode>.

knowhere-v2.3.0-beta

Release Notes

What's new?

• Sparse Index:
  • Introduced sparse vectors. Indexes SPARSE_WAND and SPARSE_INVERTED_INDEX are supported.
• GPU-based indexes:
  • Supported CAGRA, a GPU-based graph index from Nvidia.
  • Supported BRUTE_FORCE for GPU.
• Iterator:
  • Iterating approximately nearest vector is supported. A new API Interator is introduced for HNSW and IVF-series indexes.
• Multiple Data Types:
  • Knowhere now supports BF16 and FP16 as data type.

knowhere-v2.2.5

What's Changed

• Bugfix for diskann index building (#385) by @foxspy in #386

Full Changelog: v2.2.4...v2.2.5