All pull requests

Pull requests

• perf: make chunked training vocabulary projections contiguous

  #1166·Lazycat1119 opened Nov 9, 2026·

  ·

• [Draft design 22/22] Canonical training cutover and legacy cleanup

  #1165·jiapingW opened Nov 9, 2026Collaborator·

  ·

• [Draft design 21/21] MTP training migration and legacy cleanup

  #1164·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 20/21] Native target initialization and Qwen3.5 preset

  #1163·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 19/21] Pinned SGLang export and compatibility

  #1162·jiapingW opened Nov 8, 2026Collaborator·

  ·
  1

• [Draft design 18/21] Canonical publication and standalone HF export

  #1161·jiapingW opened Nov 8, 2026Collaborator·

  ·
  1

• [Draft design 17/21] Canonical training entry and resume

  #1160·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 16/21] Training objectives and input preparation

  #1159·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 15/21] Checkpoint adapters and public draft loading

  #1158·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 14/21] Checkpoint artifacts and validated weight loading

  #1157·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 13/21] DMTP configuration, fusion and model

  #1156·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 12/21] Vocabulary mappings, EAGLE3 and P-EAGLE models

  #1155·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 11/21] Canonical DSpine execution and transition scoring

  #1152·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 10/21] Canonical DSpark and Domino models

  #1149·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 09/21] Canonical DFlash, MTP and DFlash2 models

  #1146·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 08/21] Heads, fusion and model assembly

  #1141·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 07/21] Decoder composition, convolution and MLA

  #1130·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 06/21] Rotary components and attention execution

  #1124·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 05/21] Shared backbone, HF interfaces and base components

  #1116·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 04/21] EAGLE and MTP configuration codecs

  #1109·jiapingW opened Nov 8, 2026Collaborator·

  ·

• [Draft design 03/21] Legacy rules and DFlash-family configuration codecs

  #1101·jiapingW opened Nov 8, 2026Collaborator·

  ·
  1

• [Draft design 02/21] DraftConfig composition and canonical parsing

  #1093·jiapingW opened Nov 8, 2026Collaborator·

  ·
  5

• [Draft design 01/21] Component specs and codecs

  #1086·jiapingW opened Nov 8, 2026Collaborator·

  ·
  6

• feat(training): save_on_epoch_end - pinned per-epoch checkpoints exempt from rotation

  #1081·subhajitdchow opened Nov 7, 2026·3 of 6 tasks·

  ·
  1

• ci: cache flash-attn wheels only after the ABI check, trust the workspace once

  #1080·curnane-lab opened Nov 7, 2026Collaborator·4 of 6 tasks·

  ·