All pull requests

Pull requests

• Support compiled CUDA stream recording for SDM tensors

  #1295·aw471 opened Oct 29, 2026Collaborator·

  ·
  2

• Allow bounded single-chunk splits under torch.compile

  #1294·aw471 opened Oct 29, 2026Collaborator·

  ·
  2

• Add versioned documentation publishing and navigation

  #1291·AkilsuryaS opened Oct 28, 2026·

  ·
  1

• Add opt-in CUDA Graph replay for Kumo Tabular prediction

  #1290·GraysonWu opened Oct 28, 2026·

  ·
  2

• Add fine-tuning guide

  #1280·wsad1 opened Oct 27, 2026Collaborator·

  ·
  9

• Compile categorical string lookup with native PyTorch operations

  #1275·aw471 opened Oct 27, 2026Collaborator·

  ·
  2

• Record string length bounds for compiled string operations

  #1274·aw471 opened Oct 27, 2026Collaborator·

  ·
  7

• Preserve nanosecond datetime precision with legacy microsecond defaults

  #1273·alan-kumo opened Oct 27, 2026Contributor·

  ·
  2

• Replace cuDF string sorting with native PyTorch on CUDA

  #1272·aw471 opened Oct 26, 2026Collaborator·

  ·
  7

• Add native variable-length entry permutation

  #1271·aw471 opened Oct 26, 2026Collaborator·

  ·
  5

• Preserve fitted recipe execution state across compiled conditionals

  #1258·aw471 opened Oct 23, 2026Collaborator·

  ·
  3

• Make single-member category shuffle fitting traceable

  #1243·aw471 opened Oct 22, 2026Collaborator·

  ·
  4

• Restore Kumo Forecast upstream arithmetic and precision parity

  breakingbug
  #1240·devanj08 opened Oct 22, 2026Collaborator·

  ·
  1

  5

• [DO NOT REVIEW/WIP] Add KumoTabular context-sharded inference

  #1238·aw471 opened Oct 22, 2026Collaborator·

  ·
  2

• Add MatrixFormer matrix completion model

  #1203·jacobf18 opened Oct 20, 2026·

  ·
  1

• Fuse attention projection with per-head normalization

  #1200·autumnust opened Oct 20, 2026Collaborator·

  ·
  1

• Reduce redundant attention input preparation

  #1191·autumnust opened Oct 19, 2026Collaborator·4/4·

  ·
  1

• Add MaskLatent

  #1179·cedric-lorenz opened Oct 19, 2026Collaborator·4/4·

  ·
  2

• Add MaskNearEmpty

  #1178·cedric-lorenz opened Oct 19, 2026Collaborator·3/4·

  ·
  2

• Add structured missingness processors

  #1176·cedric-lorenz opened Oct 19, 2026Collaborator·2/4·

  ·
  9

• Timestral vibe

  #1171·rusty1s opened Oct 17, 2026Collaborator·

  ·
  7

• Reduce repeated table metadata preparation during inference

  #1170·autumnust opened Oct 17, 2026Collaborator·3/4·

  ·
  7

• Fuse shared-length query scaling during inference

  #1168·autumnust opened Oct 16, 2026Collaborator·2/4·

  ·
  6

  

• Add MaskRandom

  #1166·cedric-lorenz opened Oct 16, 2026Collaborator·1/4·

  ·
  7

• Make bounded KumoRelational task propagation compilable

  #1157·aw471 opened Oct 16, 2026Collaborator·

  ·
  1