v2.6.0

What's Changed

• enhance: sparse: Add support for persistency of sparse indexes by @sparknack in #1238
• feat(simd/rvv): add RVV SIMD optimization for int8_vec_inner_product, int8_vec_L2sqr, int8_vec_norm_L2sqr by @lyd1992 in #1240
• feat(simd/rvv): add RVV SIMD optimization for int8_vec batch_4 functions by @lyd1992 in #1242
• sparse: change index file header size to 32 for now by @sparknack in #1245
•  [SIMD][RVV] add RVV SIMD optimization for bf16_vec_inner_product_rvv && bf16_vec_L2sqr_rvv && bf16_vec_norm_L2sqr_rvv by @hansu2022 in #1244
• fix: sparse vector dup key error in bruteforce range search by @sparknack in #1246
• fix: sparse: all entries should use span in version 8 by @sparknack in #1251
• [SIMD][RVV] add RVV SIMD optimization for bf16_vec_inner_product_batch_4_rvv && bf16_vec_L2sqr_batch_4_rvv by @hansu2022 in #1248
• enhance: remove useless omp parallel by @foxspy in #1256
• Add datatype check for int8 by @cqy123456 in #1255
• feat(simd-rvv): Add RVV SIMD optimized bf16 patch functions for inner product and L2sqr by @lyd1992 in #1253
• [SIMD][RVV] add RVV SIMD optimization for fp16_vec_inner_product_rvv && fp16_vec_L2sqr_rvv && fp16_vec_norm_L2sqr_rvv by @hansu2022 in #1254
• feat: Extend BitsetView for 1-hop ID mapping in hierarchical index_nodes by @alwayslove2013 in #1252
• feat(simd-rvv): Add RVV SIMD optimized batch-4 bf16 patch functions by @lyd1992 in #1259
• unify bf search and search_with_buf by @alwayslove2013 in #1260
• Accelerating Data Processing with Prefetching by @wjunLu in #1263
• feat: emb_list(hnsw) by @alwayslove2013 in #1267
• [SIMD][RVV] add RVV SIMD optimization for fp16_vec_inner_product_batch_4_rvv && fp16_vec_L2sqr_batch_4_rvv by @hansu2022 in #1269
• add index_type/data_type/mmap check for emb_list_hnsw by @alwayslove2013 in #1272

New Contributors

• @hansu2022 made their first contribution in #1244
• @wjunLu made their first contribution in #1263

Full Changelog: v2.5.11...v2.6.0

v2.5.5

What's Changed

• [bug] fix IVFPQFastScan::RangeSearch() on ARM by @alexanderguzhva in #1116
• fix ARM bf16 patch functions by @alexanderguzhva in #1120
• enhance: reduce the queuing problem caused by the mixed task of growing index and sealed index by @foxspy in #1126
• fix: add use_knowhere_build_pool param for BuildAsync by @foxspy in #1128
• [skip e2e]: update cardinal version by @foxspy in #1129
• Fill -1 for invalid ids by @chasingegg in #1131
• Enable SVE Support for L2 Metric Computation in FP32 by @adarshs1310 in #969
• RaBitQ implementation by @alexanderguzhva in #1115
• Add search level metric by @chasingegg in #1139
• fix: fix gpu build by @foxspy in #1141

New Contributors

• @adarshs1310 made their first contribution in #969

Full Changelog: v2.5.4...v2.5.5

v2.5.4

What's Changed

• [skip e2e] Change CI timeout to 120min by @cydrain in #1099
• Add Int8Vector SIMD APIs for avx512/avx2/sse/ref by @cydrain in #1098
• Add benchmark_simd_qps by @cydrain in #1101
• enhance: SCANN_DVR support different refine type in train and search process by @cqy123456 in #1106
• Add neon implementation for Int8Vector by @cydrain in #1104
• Correct some distance APIs' input parameter types by @cydrain in #1109
• Rename half_precision_floating_point_xxx to xxx_typed for better readability by @cydrain in #1110
• BruteForce support Int8Vector by @cydrain in #1114
• Fix CAGRA-HNSW serialization format by @lowener in #1108

Full Changelog: v2.5.3...v2.5.4

v2.5.3

Features

• support diskann mv(cloud only) by @chasingegg in #1045
• knowhere support data view index node by @cqy123456 in #1016

Enhance

• scann support search parameter 'ensure_topk_full' by @cqy123456 in #1072

Bugfix

• fix : for all empty sparse index, PrepareMmap will early return to avoid mmapping empty file by @sparknack in #1050

Full Changelog: v2.5.2...v2.5.3

v2.5.2

Features

• add max score ratio downscaling for approximate searching on sparse vector @sparknack in #1018
• add DAAT MaxScore support for sparse vector by @sparknack in #1015
• support MV only for HNSW by @chasingegg in #1020

Improvements

• Iterator thread control optimization by @alwayslove2013 in #1008
• memory usage optimization for sparse vector by @sparknack in #1011

Full Changelog: v2.5.1...v2.5.2

v2.5.1

What's Changed

• enhance: scann support iterator by @cqy123456 in #992
• make FileReader to close fd using RAII by @zhengbuqian in #1005

Full Changelog: v2.5.0...v2.5.1

v2.5.0

draft

What's Changed

• update raft to 24.10 by @presburger in #914
• Fix group search by @alexanderguzhva in #913
• fix : add config check for open and closed range settings by @foxspy in #915
• fix: knowhere ut bugfix by @foxspy in #918
• raft index supports cosine similarity by normalizing the input data. by @presburger in #924
• compensate for the missing acceleration functions in ARM NEON. by @presburger in #922
• Allows FAISS_HNSW_FLAT index to load & search on hnswlib flat index by @alexanderguzhva in #929
• enhance : use faiss hnsw implementation to replace hnswlib by @foxspy in #930
• enable refine for FAISS HNSW indices by default, if available by @alexanderguzhva in #932
• fix: extend ut time and fix some compile warnings by @foxspy in #931
• improve sparse vector index mmap: to mmap almost everything by @zhengbuqian in #928
• fix: update faiss hnsw unit test for compatibility by @foxspy in #934
• Add SparseInvertedIndexNodeCC by @zhengbuqian in #933
• sparse mmap mode to use a disk file as backed mmap mmeory by @zhengbuqian in #935
• fix: add hnsw pq/prq refine check by @foxspy in #936
• use MAP_PRIVATE for mmapped file by @zhengbuqian in #938
• fix : fix code checker error in milvus pipeline by @foxspy in #939
• enhance : optimize config check error msg by @foxspy in #940
• enhance: BF supports ids of base_data starting from a specific value by @cqy123456 in #941
• sparse RangeSearch/AnnIterator to return raw distance by @zhengbuqian in #944
• fix : optimize faiss hnsw refine check error msg by @foxspy in #945
• fix : update knowhere index version by @foxspy in #947
• enhance : update vector index defeault value by @foxspy in #946
• fix: BitsetViewSelectorHelper miss id_offset by @cqy123456 in #948
• Add optimized distiance functions for PowerPC by @carll99 in #894
• enhance: optimize get norms function with simd and reduce some calculations by @cqy123456 in #950
• fix: Isolate faiss_hnsw and hnsw by index version by @foxspy in #952
• Update benchmark for RAFT by @cydrain in #957
• add support for android by @presburger in #959
• delete useless deps for milvus lite on knowhere by @presburger in #960
• [skip e2e] Automated Update Cardinal Commit by @sre-ci-robot in #963
• wand optimize cursor class by @zhengbuqian in #968
• optimize conanfile by @presburger in #965
• fix: conan: make glog shared by @sparknack in #972
• Update docker image for ci by @cydrain in #976
• Fix build error by @cydrain in #979
• [skip ci] Update Dockerfile to support gcc-12 better by @cydrain in #983
• [skip ci] Update Dockerfile by @cydrain in #985
• Update docker image to fix arm build error by @cydrain in #984
• enhance: BF functions support real fp16/bf16 calculate by @cqy123456 in #980
• Optimize feature registry by @cydrain in #990
• Fix some build warnings by @cydrain in #993
• Update conan pack to save UT build time by @cydrain in #994
• faiss_hnsw support INT8 by @cydrain in #991
• Enable mmap-ped Faiss indices (IndexFlatCodes-based and HNSW) by @alexanderguzhva in #996
• fix : adapt for cardinal ci by @foxspy in #999

New Contributors

• @carll99 made their first contribution in #894
• @sparknack made their first contribution in #972

Full Changelog: v2.3.12...v2.5.0

v2.3.14

What's Changed

• enhance: optimize sparse index and get ~10% performance improvement @zhengbuqian in #968

Full Changelog: v2.3.13...v2.3.14

v2.3.13

Features

• update raft to 24.10 by @presburger in #914
• raft index supports cosine similarity by @presburger in #924
• add optimized distiance functions for PowerPC by @carll99 in #894

Improvements

• accelerate the distance computation for ARM NEON by @presburger in #922
• mmap almost everything for sparse vector index by @zhengbuqian in #928
• accelerate the brute-force performance with cosine metric by adding avx simd support and pre-compute the norms by @cqy123456 in #950

New Contributors

• @carll99 made their first contribution in #894

Full Changelog: v2.3.12...v2.3.13

v2.3.12

Features

• async build support for cardinal by @presburger in #873
• provide a static checking method for index config by @foxspy in #888

Improvements

• add a bit-based structure to mark index-specific features by @foxspy in #887

Bug Fixes

• fix: copy the query vector to resolve the use-after-free issue for ivf-iterator by @alwayslove2013 in #901

Full Changelog: v2.3.11...v2.3.12