All pull requests

Pull requests

• fix(serverless): fork checkpoints across W&B entities

  #1148·Boulea7 opened Oct 26, 2026·

  ·
  1

  1

• fix: annotate tic-tac-toe self-play rollout as returning a tuple

  #1147·Avicennasis opened Oct 25, 2026·

  ·

• fix: detect CUDA 13 from the new nvidia-smi header

  #1146·alleyxd opened Oct 24, 2026·

  ·

• fix: cancel sibling rollouts when a gather fails

  #1145·salma2vec opened Oct 22, 2026·4 of 5 tasks·

  ·
  1

• fix: keep group and trajectory provenance in ruler_score_group

  #1143·salma2vec opened Oct 22, 2026·4 of 5 tasks·

  ·
  1

• fix: cancel started rollouts when trajectory_group_batches exits

  #1141·salma2vec opened Oct 22, 2026·5 of 6 tasks·

  ·
  1

• fix: reject unsupported Unsloth RL gradient accumulation

  #1137·monsterlag opened Oct 21, 2026·

  ·
  1

• Fix benchmark loading from paths containing apostrophes

  #1131·yunyv opened Oct 20, 2026·

  ·

• fix: avoid cascading benchmark display-name renames

  #1128·yunyv opened Oct 19, 2026·

  ·

• perf(unsloth): skip ignored rows in LM head projection

  #1127·rapidSpace opened Oct 19, 2026·

  ·
  1

• fix(trainer-rank): expose Adam epsilon in AdamParams

  #1126·rapidSpace opened Oct 18, 2026·

  ·
  1

• test: compare TrainerRank hidden-state paths at equal packing

  #1125·bradhilton opened Oct 15, 2026Collaborator·

  ·

• Add optional Parallel Search MCP support to MCP-RL

  #1123·georgeatparallel opened Oct 11, 2026·

  ·

• refactor: use the shared zero-variance check for native Tinker advantages

  #1122·bradhilton opened Oct 10, 2026Collaborator·

  ·

• fix(trainer-rank): release the native cache before refusing gathered outputs

  #1106·bradhilton opened Oct 9, 2026Collaborator·

  ·

• Preserve native dtype in distributed group_mean instead of casting to float32

  #1103·OnePunchMonk opened Oct 3, 2026·1 of 2 tasks·

  ·
  1

• perf: reuse MoE terms within one memory evaluation

  #1101·bradhilton opened Oct 3, 2026Collaborator·

  ·

• fix(trainer-rank): credit reusable cache when placing gathered outputs

  #1100·bradhilton opened Oct 3, 2026Collaborator·

  ·

• fix: sum nonfused column LoRA input gradients

  #1092·bradhilton opened Oct 3, 2026Collaborator·

  ·
  1

• fix(trainer-rank): price dense CP2 recompute and no-grad waves from traced stages

  #1090·bradhilton opened Oct 3, 2026Collaborator·

  ·

• fix: preserve input gradients in fused LoRA layers

  #1087·bradhilton opened Oct 3, 2026Collaborator·

  ·

• perf(megatron): reuse packed weights for frozen MoE experts

  #1061·bradhilton opened Oct 2, 2026Collaborator·

  ·

• perf(trainer-rank): try budgeted cache recovery before splitting

  #1053·bradhilton opened Sep 30, 2026Collaborator·

  ·
  2

• fix: preserve container state and dictionary order during compaction

  #1015·bradhilton opened Sep 28, 2026Collaborator·

  ·

• fix(ruler): add score normalization, clamping, and robust JSON extraction (#477)

  #1004·shobhitagnihotri69 opened Sep 28, 2026·

  ·
  1