All pull requests

Pull requests

• docs: explain RAG data preparation and context budgets

  documentation
  #1221·rudrakshkarpe opened Oct 9, 2026·3 of 4 tasks·

  ·
  1

  2

• fix(evaluate): support throughput and sweep baselines without speculative decoding

  dco-faileddocumentation
  #1220·chinmay29 opened Oct 9, 2026·4 tasks done·

  ·
  1

  6

• [Bugfix] Make offline data generation robust to truncated files

  documentation
  #1219·DarkLight1337 opened Oct 9, 2026Member·4 tasks done·

  ·
  3

• fix(train): resolve uncached verifier IDs in offline draft loading

  documentation
  #1218·Liuchenbing-2026 opened Oct 9, 2026·3 of 4 tasks·

  ·
  2

• Add strict DFlash2 selector miss policy

  dco-failedquality-failed
  #1217·shubhra opened Oct 9, 2026Collaborator·4 tasks·

  ·
  4

• feat(cli): show resume progress and cumulative counters in regenerate…

  #1215·ababyshark opened Oct 9, 2026·3 tasks done·

  ·
  6

• fix(models): use Gemma-style final norm for Qwen3.5 MoE

  #1213·Liuchenbing-2026 opened Oct 9, 2026·2 of 4 tasks·

  ·
  3

• Add OTEL trace replay to evaluate.py

  dco-faileddocumentation
  #1208·reed-meyerson opened Oct 8, 2026Contributor·4 tasks done·

  ·
  3

• Add a throughput vs. interactivity benchmark (InferenceX-style) with a DSpark and MTP example

  documentation
  #1204·eldarkurtic opened Oct 7, 2026Collaborator·

  ·
  16

• Support gradient checkpointing for EAGLE-3 with flex attention

  #1203·erfansoheil opened Oct 7, 2026·3 of 4 tasks·

  ·
  5

• Add flex attention kernel options for low shared-memory GPUs (sm120, sm89)

  #1202·erfansoheil opened Oct 7, 2026·3 of 4 tasks·

  ·
  5

• docs: Link dashboard and remove supported models table

  documentationready
  #1201·rahul-tuli opened Oct 7, 2026Collaborator·3 of 4 tasks·

  ·
  6

  

• Add unit tests, type annotations, and docstrings for train utilities (fixes #1092)

  #1200·jasonchen505 opened Oct 7, 2026·4 tasks done·

  ·
  1

  6

• fix(data): abort dataset build when the render endpoint fails

  #1196·li-lizhe opened Oct 5, 2026·

  ·
  1

  4

• [Refactor][Data] Share PreparedSample and remove local processors [2/n]

  documentation
  #1195·WindChimeRan opened Oct 5, 2026Contributor·3 of 4 tasks·

  ·
  2

• Enable gradient accumulation

  documentationneeds-rebaseready
  #1178·shanjiaz opened Sep 30, 2026Collaborator·4 tasks done·

  ·
  6

• [feat]: add DeepSeek V4 aliases

  #1177·pavelgein opened Sep 30, 2026Contributor·2 of 4 tasks·

  ·
  3

• [Train] Add --draft-num-key-value-heads; warn on MLA verifiers

  documentation
  #1176·Frozen7771 opened Sep 30, 2026Contributor·4 tasks done·

  ·
  1

  5

• Add e2e, regression, and multi-GPU tests to Buildkite nightly pipeline

  ready
  #1161·rashmigottipati opened Sep 24, 2026Collaborator·4 tasks·

  ·
  11

• fix(models): treat the Qwen3.5 MoE family as Gemma-style for the veri…

  #1145·avshalomman opened Sep 18, 2026·4 tasks·

  ·
  2

• fix(eagle3): fail conversion on unmatched checkpoint keys instead of saving random weights

  #1131·YuEfSaEDU opened Sep 16, 2026·

  ·
  1

  2

• Loss weight schedule

  documentationready
  #1130·fynnsu opened Sep 16, 2026Collaborator·3 of 4 tasks·

  ·
  5

• feat(benchmark): add 'comment' subcommand for GitHub PR summaries

  #1125·orestis-z opened Sep 15, 2026Collaborator·4/4·4 tasks done·

  ·
  7

  

• feat(benchmark): add HTML and Markdown report generation

  needs-rebase
  #1124·orestis-z opened Sep 15, 2026Collaborator·3/4·4 tasks done·

  ·
  8

  

• feat(benchmark): add torch.profiler support and GPU kernel breakdown

  #1123·orestis-z opened Sep 15, 2026Collaborator·2/4·4 tasks done·

  ·
  6