All pull requests

Pull requests

• [TimesFM3 12/n] Preprocess model inputs

  #1019·guerrantif opened Sep 28, 2026Collaborator·12/12·

  ·
  1

• [TimesFM3 11/n] Assemble model layers

  #1018·guerrantif opened Sep 28, 2026Collaborator·11/12·

  ·
  1

• Chunk recipe execution and output processing

  #1017·ValterH opened Sep 28, 2026Collaborator·

  ·
  1

• Chunk expensive numerical operations

  #1016·ValterH opened Sep 28, 2026Collaborator·1/2·

  ·
  1

• Chunk Kumo row embeddings

  #1015·ValterH opened Sep 28, 2026Collaborator·2/2·

  ·
  2

• Reduce numerical preprocessing temporaries

  #1014·ValterH opened Sep 28, 2026Collaborator·

  ·
  2

• Share chunk-memory sizing

  #1013·ValterH opened Sep 28, 2026Collaborator·1/2·

  ·
  2

• Avoid ensemble selection copies

  #1012·ValterH opened Sep 28, 2026Collaborator·

  ·
  2

• create an example under examples/relational/finetune.py

  #1011·alan-kumo opened Sep 28, 2026·

  ·
  5

• Add Kumo-Anomaly diffusion denoiser components

  #1010·agautam478 opened Sep 28, 2026Contributor·

  ·
  1

• Use automatic estimator batching by default

  #1009·RBendias opened Sep 28, 2026Collaborator·

  ·
  1

• Use automatic batching in tabular benchmarks

  #1008·RBendias opened Sep 28, 2026Collaborator·

  ·
  1

• Add automatic estimator batching

  #1007·RBendias opened Sep 28, 2026Collaborator·

  ·
  2

• Cast chunked attention outputs to the destination dtype

  #1004·RBendias opened Sep 28, 2026Collaborator·

  ·
  4

  

• [TimesFM3 10/n] Add CPM RevIN refinement

  #1001·guerrantif opened Sep 28, 2026Collaborator·10/12·

  ·
  1

• [TimesFM3 9/n] Add transformer stack

  #999·guerrantif opened Sep 27, 2026Collaborator·9/12·

  ·
  1

• Replay CUDA graphs where they are measured to be worth it

  #995·JingangQu opened Sep 27, 2026Collaborator·

  ·
  1

• Add RankGaussian numerical preprocessor to Kumo Tabular

  #993·ValterH opened Sep 26, 2026Collaborator·2/2·

  ·
  3

• [TimesFM3 8/n] Add mixing transformer

  #969·guerrantif opened Sep 24, 2026Collaborator·8/12·

  ·
  1

• Fix QuantileTransform for 3D inputs

  #964·rusty1s opened Sep 24, 2026Collaborator·

  ·
  4

• Enable codecov

  #958·akihironitta opened Sep 23, 2026Member·

  ·
  3

• [TimesFM3 7/n] Add multi-head attention

  #955·guerrantif opened Sep 23, 2026Collaborator·7/12·

  ·
  1

• [TimesFM3 6/n] Add attention mask and rotary positional embeddings

  #952·guerrantif opened Sep 23, 2026Collaborator·6/12·

  ·
  2

• Support INT8 storage for KV cache

  #950·akihironitta opened Sep 23, 2026Member·

  ·
  2

• [TimesFM3 5/n] Add attention scaling

  #946·guerrantif opened Sep 22, 2026Collaborator·5/12·

  ·
  2