All pull requests

Pull requests

• perf(qwen4exp): faster two-GPU split decode (overlap, fewer launches, draft)

  #937·davide221 opened Nov 4, 2026Contributor·6 of 12 tasks·

  ·

• perf(deepseek41): two-box decode switches in ds41-r9700-gorgon

  #936·davide221 opened Nov 3, 2026Contributor·

  ·
  7

• fix(qwen4exp): keep MTP drafting while a thinking budget is open

  #935·davide221 opened Nov 3, 2026Contributor·

  ·
  3

• feat(kvflash): exact attention-mass chunk scorers on CPU and GPU, user-question query helpers, mode 5 drafter scorer

  #934·dusterbloom opened Nov 3, 2026Collaborator·

  ·
  28

• perf(qwen4exp): faster split mode on R9700 + Strix Halo

  #933·davide221 opened Nov 3, 2026Contributor·

  ·
  2

• feat(kvflash): pager slices, step fences, deferred free and the KVFLASH2 blob

  #932·dusterbloom opened Nov 3, 2026Collaborator·

  ·
  11

• fix(examples): support Windows stream pipe handle inheritance in chat example

  #931·hopstreax opened Nov 3, 2026Contributor·

  ·
  1

  4

• fix(server): the built-in Qwen renderer writes an empty thinking block where the template does (stacked on #898)

  #900·Graffioh opened Oct 28, 2026Collaborator·

  ·

• fix(server): the built-in Qwen renderer writes one tool per line, as the template does (stacked on #898)

  #899·Graffioh opened Oct 28, 2026Collaborator·

  ·

• fix(server): the built-in Qwen renderer follows the model's template for past thinking (stacked on #896)

  #898·Graffioh opened Oct 28, 2026Collaborator·

  ·

• fix(gemma4): preserve sliding-window KV history during verification

  #824·pculaf opened Oct 7, 2026Contributor·

  ·

• perf(qwen4exp): enable guarded GPU argmax on gfx1151

  #823·dusterbloom opened Oct 7, 2026Collaborator·

  ·

• feat(vulkan): add native LFM2.5 MoE alongside Qwen

  #807·Amidwestnoob opened Oct 3, 2026Contributor·

  ·
  18

• Add JEV compatible interface

  #790·howard0su opened Sep 30, 2026Contributor·

  ·
  9

• feat(profiling): emit lucegraph concurrency capture in-tree

  #783·Graffioh opened Sep 29, 2026Collaborator·

  ·

• ci(e2e): run Qwen and DS4 end to end on lucebox3

  #756·Graffioh opened Sep 23, 2026Collaborator·4 tasks·

  ·
  24

• feat(pflash): auto skip-park residency with VRAM estimate and OOM fallback

  #747·Graffioh opened Sep 20, 2026Collaborator·2 tasks done·

  ·

• feat(server): opt-in logprobs on chat completions (AR path, qwen35 + ds4)

  #737·Graffioh opened Sep 16, 2026Collaborator·5 of 6 tasks·

  ·

• feat(server): carry drafter feature ring in concurrent prefix checkpoints

  #730·Graffioh opened Sep 14, 2026Collaborator·

  ·
  12

• refactor(test): organize component suites and replace simulated coverage

  #709·Graffioh opened Sep 5, 2026Collaborator·

  ·
  34

• feat(pflash): improve AMD defaults and enable rocWMMA

  #669·Graffioh opened Aug 27, 2026Collaborator·

  ·

• experimental: integrate luce_odistill bridge

  #663·Graffioh opened Aug 26, 2026Collaborator·

  ·

• feat(test): add manual Qwen 3.6 production qualification

  #636·Graffioh opened Aug 20, 2026Collaborator·

  ·
  135

• fix(qwen35): size mixed concurrent target-step graphs

  #622·Graffioh opened Aug 18, 2026Collaborator·

  ·

• feat(harness): add interactive modes and GPU selection

  #611·Graffioh opened Aug 15, 2026Collaborator·2 tasks·

  ·
  5