All pull requests

Pull requests

• Add DeterministicNNDescent, a KNN graph build that does not depend on the thread count, and faster e-means sums

  CLA Signedmeta-exported
  #5849·mnorris11 opened Nov 5, 2026Contributor·

  ·
  1

• e-means: rename niter to nepochs

  CLA Signed
  #5848·swannbessa opened Nov 5, 2026Contributor·

  ·

• Flat search: use the database-parallel path for nq >= threads below the BLAS threshold

  CLA Signedmeta-exported
  #5847·weixianghong opened Nov 5, 2026·

  ·
  1

• e-means: add benchmarks reproducing the paper

  CLA Signed
  #5845·swannbessa opened Nov 5, 2026Contributor·

  ·
  1

• Use BF16 defaults and share quantized rows across E-means sharding

  CLA Signedmeta-exported
  #5844·algoriddle opened Nov 5, 2026Contributor·

  ·
  1

• Update SVS to 0.5.1

  CLA Signed
  #5843·ethanglaser opened Nov 4, 2026Contributor·

  ·
  1

• Replace the e-means README with a user guide

  CLA Signed
  #5842·swannbessa opened Nov 4, 2026Contributor·

  ·
  7

• e-means: faiss-style verbose output that reports epochs and asks for a citation

  CLA Signed
  #5841·swannbessa opened Nov 4, 2026Contributor·

  ·
  6

• fix: compact lo_codes in IndexSQFastScan::remove_ids

  #5836·sunhouy opened Nov 4, 2026·

  ·
  1

  1

• Separate Emeans implementations

  CLA Signedmeta-exported
  #5832·mdouze opened Nov 3, 2026Contributor·

  ·
  1

• fix: respect n0 in Torch reconstruct_n default count

  CLA Signed
  #5826·DiadkoShmek opened Oct 31, 2026·

  ·
  2

• Document Canberra, BrayCurtis and JensenShannon metrics

  CLA Signed
  #5825·DucAnhValentinoNguyen opened Oct 31, 2026·

  ·
  1

• Add NEON kernels for fvec_L2sqr_ny / fvec_inner_products_ny

  #5781·shiyu-zhang opened Oct 26, 2026·

  ·
  1

• Fix IndexBinaryMultiHash serialization when a bucket holds all ntotal ids

  CLA Signed
  #5780·ViderStar opened Oct 25, 2026·

  ·
  3

• Use a reservoir for large-k IVF search

  CLA Signed
  #5779·Tickdack opened Oct 25, 2026Contributor·4 of 5 tasks·

  ·
  1

• #5756 :: Forward and translate ID selectors through Refine and PreTransform

  CLA Signed
  #5778·ricardoleal20 opened Oct 25, 2026·

  ·
  1

• Move compiler-specific macros to their own section

  CLA Signed
  #5777·JAicewizard opened Oct 24, 2026Contributor·

  ·
  1

  4

• Add missing bounds checks in reconstruct and ArrayInvertedLists accessors

  CLA Signed
  #5776·Aaryan-Dadu opened Oct 24, 2026·

  ·
  1

  4

• Add BBQ layout to the CAGRA Dataset API

  CLA Signed
  #5769·tarang-jain opened Oct 22, 2026Contributor·

  ·

• Add RISC-V RVV kernels for approx top-k, Hamming top-k and uint16 partitioning

  CLA Signed
  #5767·anthony-zy opened Oct 22, 2026Contributor·

  ·

• Add RISC-V RVV kernels for FastScan

  CLA Signed
  #5765·anthony-zy opened Oct 22, 2026Contributor·

  ·

• Make HadamardRotation depth configurable (1-3 rounds)

  CLA Signed
  #5760·ranbenbasat opened Oct 20, 2026Contributor·

  ·

• Compute the coarse distances in search_preassigned when they are not given

  CLA Signed
  #5748·Kayvan-Zahiri opened Oct 16, 2026Contributor·

  ·
  1

• Add IndexIVFRaBitQSSD: IVF-RaBitQ with codes split across RAM and SSD

  CLA Signed
  #5746·YuexuanXu7 opened Oct 16, 2026·

  ·
  2

• Add dual FMA accumulator to AVX2 scalar-quantizer distance loop

  CLA Signedmeta-exported
  #5745·yfeldblum opened Oct 16, 2026Contributor·

  ·
  1