All pull requests

Pull requests

• api: refuse unsupported OpenAI result shaping

  #1787·benmaster82 opened Sep 28, 2026Contributor·

  ·

• fix(serve): an image_url that is not an object answered 500

  #1786·kevin9327 opened Sep 28, 2026Contributor·

  ·

• fix(serve): a malformed tools or messages answered 500 because rendering ran before validation

  #1785·kevin9327 opened Sep 28, 2026Contributor·

  ·

• qwen36: the DeltaNet layer runs on the card (Q36_DN_GPU=1); expert homes per layer range (QT_HOME=layer); shexp offered on request

  #1784·kreuzzelg opened Sep 28, 2026Contributor·

  ·

• fix(serve): a text part whose text is not a string answered 500

  #1783·kevin9327 opened Sep 28, 2026Contributor·

  ·

• fix(serve): a past tool call whose function is not an object answered 500

  #1782·kevin9327 opened Sep 28, 2026Contributor·

  ·

• glm53: expert workers for GLM-5.3 (CLUSTER_WORKERS / EXPERT_WORKER)

  #1774·JackKnifeAI opened Sep 27, 2026·

  ·

• qwen36 tier: re-plan the VRAM residents from the prompt's own prefill routing (QT_PREFILL_REPLAN=1)

  #1769·kreuzzelg opened Sep 27, 2026Contributor·

  ·

• feat(dsv4): add automatic multi-GPU residency and faster CUDA execution

  #1766·ZacharyZcR opened Sep 27, 2026Contributor·

  ·

• fix(api): keep GLM-5.3 reasoning out of the /v1/messages answer

  #1765·SulimanAbdulrazzaq opened Sep 26, 2026Contributor·

  ·

• tests: link $(VK_OBJ) into the rest of the engine-including rules

  #1763·Kenneth-Javier opened Sep 26, 2026Contributor·

  ·

• quant: NEON arm for batched matmul_fp8 (bit-exact, 6–10× at prefill shapes)

  #1761·mfethe1 opened Sep 26, 2026Contributor·

  ·
  1

• build: generate header prerequisites with -MMD -MP (#1741)

  #1758·Kenneth-Javier opened Sep 26, 2026Contributor·

  ·
  5

• docs: add AMD Radeon ROCm/HIP build path (community recipe)

  #1756·A1M918 opened Sep 26, 2026·

  ·
  1

• feat(qwen36): derive the default expert-cache cap from host RAM instead of a hardcoded 16

  #1747·jtinbergen opened Sep 25, 2026Contributor·

  ·

• xdna: run the lane on Linux (dlopen loader, libcoli_xdna.so helper)

  #1746·huppiflupp opened Sep 25, 2026Contributor·

  ·

• security(glm53): refuse f32 tensors shorter than the config reads

  #1740·namespaceMarcello opened Sep 24, 2026Contributor·

  ·

• security: carry the published advisory fixes to glm53, qwen36 and deepseek_v41

  #1739·namespaceMarcello opened Sep 24, 2026Contributor·

  ·
  2

• api: add Qwen3.6 tool calling support

  #1723·GenericRikka opened Sep 23, 2026Contributor·

  ·
  3

• cuda: decode fmt=9 (bf16) in the dense matmul

  #1620·DebugSultan opened Sep 19, 2026Contributor·

  ·
  2

• fix(olmoe): use the shared RAM probe, add a plausibility floor, warn …

  #1611·Cometbuster4969 opened Sep 18, 2026·

  ·
  1

  3

• Add optional desktop launcher for Windows and Linux

  #1607·ipechman opened Sep 18, 2026·

  ·

• dense Qwen3 family: engine, int4-gs64 converter, tiny-fixture oracle, registry entry

  #1578·Xore opened Sep 16, 2026·

  ·

• perf(olmoe): O(1) LRU victim selection via intrusive recency lists (#1050)

  #1571·Petsku01 opened Sep 16, 2026Contributor·

  ·
  2

• qt expert tier + expert-cache RAM ceilings (glm53 & qwen36), A3 soak-verified

  #1564·Xore opened Sep 16, 2026·

  ·