Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit History

Commits on Jun 10, 2026

• ci(bench): default subset size 50k for hnswlib parity

  Show description for 68321af
  polaz

  committed
  68321af

  Copy full SHA for 68321af

Commits on Jun 8, 2026

• ci(bench): skip quant sweep on dim greater than 4096

  Show description for ece7816
  polaz

  committed
  ece7816

  Copy full SHA for ece7816

• build: split benches/vector-ann into standalone workspace

  Show description for 5fe26ea
  polaz

  committed
  5fe26ea

  Copy full SHA for 5fe26ea

• ci(bench): add rabitq+sq8 quantization sweep to high-D step

  Show description for 994d9b2
  polaz

  committed
  994d9b2

  Copy full SHA for 994d9b2

• ci(bench): bump high-D ef sweep upper to 1600

  Show description for 5d11eb8
  polaz

  committed
  5d11eb8

  Copy full SHA for 5d11eb8

• ci(bench): skip smoke when high_d_dimensions specified

  Show description for f9127ba
  polaz

  committed
  f9127ba

  Copy full SHA for f9127ba

• ci(bench): add ef=800 to high-D sweep for recall floor 0.95

  Show description for 5804bd6
  polaz

  committed
  5804bd6

  Copy full SHA for 5804bd6

• ci(bench): cap high-D tiers at 20k subset to fit 30 min timeout

  Show description for 8635fbc
  polaz

  committed
  8635fbc

  Copy full SHA for 8635fbc

Commits on Jun 7, 2026

• ci(bench): drop competitor step, CN-only in CI

  polaz

  committed
  e443cbc

  Copy full SHA for e443cbc

• ci(bench): drop docker compose build, require pre-built images

  polaz

  committed
  ca9b96c

  Copy full SHA for ca9b96c

• ci(bench): subset 20k, timeout 30 min

  polaz

  committed
  5c5f366

  Copy full SHA for 5c5f366

• ci(bench): hard 20-min ceiling + 50k subset cap, keep cell surface

  polaz

  committed
  9722f1d

  Copy full SHA for 9722f1d

• Revert "ci(bench): hard-lock 20-min ceiling, drop full_sweep marketing matrix"

  Show description for 800d0f6
  polaz

  committed
  800d0f6

  Copy full SHA for 800d0f6

• ci(bench): hard-lock 20-min ceiling, drop full_sweep marketing matrix

  polaz

  committed
  de02fdc

  Copy full SHA for de02fdc

• ci(bench): copy entire docker/ tree to /tmp/compose so build context resolves

  polaz

  committed
  e0b8f7b

  Copy full SHA for e0b8f7b

Commits on Jun 6, 2026

• ci(bench): qdrant high-D coverage in competitor step

  Show description for 09d2a9c
  polaz

  committed
  09d2a9c

  Copy full SHA for 09d2a9c

• ci(bench): high_d_dimensions input for D-tier sweep

  Show description for c72eb6a
  polaz

  committed
  c72eb6a

  Copy full SHA for c72eb6a

• ci(bench): workflow_dispatch can run competitors + high-D probe

  Show description for 3614079
  polaz

  committed
  3614079

  Copy full SHA for 3614079

• ci(bench): add SIFT M=24 cell as M-scaling regression detector

  Show description for 15e401a
  polaz

  committed
  15e401a

  Copy full SHA for 15e401a

• ci(bench): gate exact baseline cells behind full sweep

  Show description for 8b62d05
  polaz

  committed
  8b62d05

  Copy full SHA for 8b62d05

• ci(bench): add exact recall=1.0 baseline cells to lean push

  Show description for d3b2fb9
  polaz

  committed
  d3b2fb9

  Copy full SHA for d3b2fb9

• ci(bench): lean push gate (20 min budget), full matrix manual-only

  Show description for 14ccf5c
  polaz

  committed
  14ccf5c

  Copy full SHA for 14ccf5c

• ci(bench): revert ef sweep to 6-point baseline

  Show description for 540f204
  polaz

  committed
  540f204

  Copy full SHA for 540f204

• ci(bench): drop ef=1600/2000 to fit 3h timeout

  Show description for 2ff52e3
  polaz

  committed
  2ff52e3

  Copy full SHA for 2ff52e3

Commits on Jun 5, 2026

• ci(bench): extend ef sweep to recall=0.998+ tail

  Show description for 83ad4a6
  polaz

  committed
  83ad4a6

  Copy full SHA for 83ad4a6

Commits on Jun 4, 2026

• bench(vector): real search-MT via --search-threads N (E_core unblocker)

  Show description for c7fa997
  polaz

  committed
  c7fa997

  Copy full SHA for c7fa997

• bench(vector): 100k subset by default + ST+MT4 sweep per push

  Show description for ebb5519
  polaz

  committed
  ebb5519

  Copy full SHA for ebb5519

• ci(bench): add EoS oversample=2.5 sweet-spot probe

  Show description for e90f987
  polaz

  committed
  e90f987

  Copy full SHA for e90f987

• ci(bench): trim glove RaBitQ to 2 oversample rows (5h timeout fix)

  Show description for 06421a5
  polaz

  committed
  06421a5

  Copy full SHA for 06421a5

Commits on Jun 3, 2026

• feat(vector): RaBitQ EndOfSearch oversampling knob

  Show description for d67fc2d
  polaz

  committed
  d67fc2d

  Copy full SHA for d67fc2d

• feat(vector): RerankMode knob (Inline | EndOfSearch | None) for RaBitQ

  Show description for 5248f69
  polaz

  committed
  5248f69

  Copy full SHA for 5248f69

• ci(bench): add glove RaBitQ α=1.2 sweep alongside legacy α=1.0

  Show description for b1004cd
  polaz

  committed
  b1004cd

  Copy full SHA for b1004cd

Commits on Jun 1, 2026

• ci(bench): thread-budget-fair sweep matrix + ann-benchmarks adapter wiring

  Show description for 8d1f4ed
  polaz

  committed
  8d1f4ed

  Copy full SHA for 8d1f4ed

• test(vector): isolate RaBitQ recall bug + cap rayon to 4 threads in CI

  Show description for 3381b6d
  polaz

  committed
  3381b6d

  Copy full SHA for 3381b6d

• ci(bench): add SIFT-128 RaBitQ diagnostic to isolate padding bug

  Show description for 3930663
  polaz

  committed
  3930663

  Copy full SHA for 3930663