All pull requests

Pull requests

• feat: make upload tags configurable

  #506·officialayush5839-arch opened Oct 19, 2026·1 of 2 tasks·

  ·
  10

• feat: add streaming ARA and backward restoration

  #503·strangecreator opened Oct 17, 2026·2 tasks·

  ·
  3

• fix: keep CoT-skip recheck from appending reasoning text to the response prefix

  #501·akushonkamen opened Oct 17, 2026·

  ·
  1

  4

• feat: Refusal classifier scorer plugin

  #500·Vinay-Umrethe opened Oct 16, 2026Collaborator·2 tasks·

  ·
  1

  7

• perf(ara): compute closure outputs via low-rank update instead of dense weight

  #499·varunbln opened Oct 16, 2026·

  ·
  13

• fix: support models whose chat template rejects the system role

  #498·varunbln opened Oct 16, 2026·

  ·
  8

• fix: handle weight files without LFS info when reading hashes

  #497·Jacob-9909 opened Oct 16, 2026·

  ·

• fix: score top trials on merged models to close live-LoRA/export gap

  #495·akushonkamen opened Oct 14, 2026·

  ·
  1

  1

• fix: remember the Hugging Face token between uploads

  #492·Yi-111-a opened Oct 9, 2026·

  ·
  1

• fix: point the reproduction README SHA256SUMS link at the Hugging Face blob

  #479·mvanhorn opened Oct 4, 2026·

  ·
  1

  1

• fix: render concurrent Rich progress bars correctly

  #457·umran666 opened Sep 27, 2026Contributor·

  ·
  99

• feat: add response logger plugins

  #451·AlisinaDevelo opened Sep 18, 2026·

  ·
  7

• feat: model tensor integrity check

  #447·hpnyaggerman opened Sep 6, 2026·

  ·
  9

• fix: stop retrying non-recoverable dtype errors

  #435·xuyua9 opened Aug 27, 2026·

  ·
  1

• fix: improve dtype fallback error reporting

  #434·xuyua9 opened Aug 27, 2026·

  ·
  1

• feat: reproduce local datasets from a Hub source

  #433·aimansour opened Aug 24, 2026·

  ·
  2

• feat(ara): add DiffusionGemma block-diffusion model support

  #421·umran666 opened Aug 3, 2026Contributor·

  ·
  1

• feat: llm as a judge for binary classification

  #419·red40maxxer opened Jul 23, 2026Contributor·

  ·
  9

• fix: reject empty scorer prompt sets

  #408·Bortlesboat opened Jul 9, 2026Contributor·

  ·
  4

• feat: add support for Qwen3Next shared experts

  #391·umran666 opened Jun 18, 2026Contributor·

  ·
  3

• Use adaptive per-method batch sizing, drop startup benchmark

  #390·kali113 opened Jun 17, 2026Contributor·

  ·
  1

  17

• feat: generate model card on local save (#258)

  #388·umran666 opened Jun 17, 2026Contributor·

  ·
  2

• fix: capture residual hidden states via forward hooks instead of output_hidden_states

  #384·kali113 opened Jun 16, 2026Contributor·

  ·
  1

  9

• feat: diffusion LLM abliteration support (DiffusionGemma)

  #378·anurag12-webster opened Jun 13, 2026·

  ·
  1

  22

• fix: preserve Hugging Face token in session to avoid repeated prompt (#142)

  #366·umran666 opened Jun 10, 2026Contributor·

  ·
  7