All pull requests

Pull requests

• Raise when a configured checkpoint load path does not exist

  #252·mayuriphad opened Oct 13, 2026·3 of 4 tasks·

  ·
  1

  1

• Seek over keyframe groups that hold no sampled frame

  coremedium
  #235·amazloumi opened Oct 9, 2026Member·10 tasks done·

  ·
  1

• Select video frames relative to the stream's first frame

  corehigh
  #233·amazloumi opened Oct 8, 2026Member·11 tasks done·

  ·
  1

  1

• Add a video config on a pretrained 0.6B backbone to the VLM example

  medium
  #232·amazloumi opened Oct 8, 2026Member·9 of 11 tasks·

  ·

• Resume when a trainable parameter has never received a gradient

  blockingcore
  #231·amazloumi opened Oct 8, 2026Member·14 tasks done·

  ·
  1

  1

• Use model.norm_eps for the query/key and cross-attention block norms

  corehigh
  #230·amazloumi opened Oct 8, 2026Member·9 tasks done·

  ·
  1

  1

• Add a converter from a Hugging Face backbone to a VLM warm-start checkpoint

  medium
  #229·amazloumi opened Oct 8, 2026Member·13 of 15 tasks·

  ·
  1

  1

• Reorganize examples/vlm into one flat layout

  low
  #227·amazloumi opened Oct 7, 2026Member·4 of 6 tasks·

  ·
  1

  1

• Decouple core and its docs from examples/

  corelow
  #226·amazloumi opened Oct 7, 2026Member·4 of 6 tasks·

  ·
  1

  1

• Document where contributions belong in CONTRIBUTING.md

  low
  #225·amazloumi opened Oct 7, 2026Member·4 of 6 tasks·

  ·
  1

  1

• Run examples/vlm/tests in the example CI workflow

  medium
  #224·amazloumi opened Oct 7, 2026Member·4 of 6 tasks·

  ·
  1

  1

• Support sequence packing under pipeline parallelism

  coremedium
  #208·timothyngo opened Sep 21, 2026Collaborator·3 tasks done·

  ·
  5

• Validate and benchmark the FlexAttention packed path

  low
  #207·timothyngo opened Sep 20, 2026Collaborator·5 tasks done·

  ·
  2

• Write the emergency checkpoint when dataloader workers are in use

  blockingcore
  #203·Naeemkh opened Sep 14, 2026Member·3 of 6 tasks·

  ·
  7

• Decouple attention head_dim from dim // n_heads

  coremedium
  #198·amazloumi opened Sep 2, 2026Member·13 tasks done·

  ·
  1

  4

• Cache visual features in the VLM eval decode loop

  #166·camilobrownpinilla opened Jul 23, 2026Collaborator·4 tasks done·

  ·
  1

  1

• Generalize VLM modality strategies; add encode_visual seam

  #165·camilobrownpinilla opened Jul 23, 2026Collaborator·4 tasks done·

  ·
  1

  3

• Feat/eval tracking support

  core
  #162·camilobrownpinilla opened Jul 23, 2026Collaborator·4 tasks done·

  ·
  1

  4

• Feat/multi gpu eval

  #160·camilobrownpinilla opened Jul 22, 2026Collaborator·4 tasks done·

  ·
  1

• Feat/text multi image eval

  core
  #148·camilobrownpinilla opened Jul 16, 2026Collaborator·4 tasks done·

  ·
  1

  1