All pull requests

Pull requests

• [BugFix][KV Cache] Avoid false Mamba external-state headroom within the same block

  module:tests
  #19940·KeyanHu-git opened Oct 27, 2026Contributor·3 of 5 tasks·

  ·
  2

• [Test][PCP] Align MLA preprocess assertions with gate output

  module:testsready-precise
  #19938·recky-c opened Oct 27, 2026Contributor·

  ·
  2

• [CI] Remove redundant OBS artifact uploads

  ci/build
  #19937·MrZ20 opened Oct 27, 2026Contributor·5 tasks·

  ·
  2

• [BugFix][xlite] Fix xlite fetching wrong cached_lens from upstream during spec decode due to host-side optimistic metadata update

  #19936·SijieFu opened Oct 27, 2026Contributor·5 tasks·

  ·
  4

• [CI] modify model path in Qwen3.6-27B-W8A8-310p-300I-DUO.yaml

  module:tests
  #19935·guxin108 opened Oct 27, 2026Contributor·

  ·
  3

• fix(ops): reject non-contiguous inputs in muls_add_triton and expand boundary coverage

  module:opsmodule:tests
  #19932·Cold-E opened Oct 27, 2026·4 of 5 tasks·

  ·
  6

• [Doc][Misc]fix Qwen3.5-397B benchmark command

  documentation
  #19931·Kunkun066 opened Oct 27, 2026·

  ·
  4

• [CI] modify model path in Qwen3.6-27B-W8A8-310p-300I-DUO.yaml

  module:tests
  #19928·guxin108 opened Oct 27, 2026Contributor·

  ·
  5

• [Doc]Add “huawei” before the specified “ascend” text.

  documentation
  #19927·Xuyzhen opened Oct 27, 2026Contributor·2 of 5 tasks·

  ·
  6

• [Perf][DSv4.1][eager] Trim the last KV-source layer in decoder replay

  module:tests
  #19926·wxsIcey opened Oct 27, 2026Collaborator·

  ·
  4

• [Doc][Misc]fix Qwen3.5-397B benchmark command

  documentation
  #19925·Kunkun066 opened Oct 27, 2026·

  ·
  4

• [Doc][Misc] Clarify cudagraph_capture_sizes and update Qwen3.8-27B hardware requirements

  documentation
  #19924·AJF-cmd opened Oct 27, 2026Contributor·5 tasks done·

  ·
  5

• [BugFix][Config] Treat a fine-grained TP size of 1 as disabled

  documentationmodule:coremodule:tests
  #19923·18391202266 opened Oct 27, 2026·5 tasks done·

  ·
  1

  4

• [Doc] Clarify Qwen3.5 PD service startup steps for v0.30.0

  documentation
  #19922·kitaikuyo421esfj-hub opened Oct 27, 2026Contributor·5 tasks·

  ·
  4

• [Doc] skill modify

  documentation
  #19921·herizhen opened Oct 27, 2026Contributor·

  ·
  5

• [Doc] Clarify Qwen3.5 PD service startup steps

  documentation
  #19920·kitaikuyo421esfj-hub opened Oct 27, 2026Contributor·5 tasks·

  ·
  4

• [BugFix] Run ENCODER_ONLY layers bidirectionally in pooling models

  module:tests
  #19919·do420 opened Oct 27, 2026Contributor·2 of 5 tasks·

  ·
  1

  2

• [BugFix] Strip stream_options on layerwise proxy decode path when stream is false

  #19917·wuming2027 opened Oct 27, 2026·

  ·
  1

  2

• [BugFix][EPLB] Complete MRv1 AllGather redundant expert routing and migration

  documentationmodule:opsmodule:tests
  #19916·Sweet-BeanOnE opened Oct 27, 2026·4 of 5 tasks·

  ·
  1

  9

• Qwen38 mtp qsa draft layer 2

  documentationmodule:coremodule:opsmodule:quantizationmodule:tests
  #19914·sunny-rain-63 opened Oct 27, 2026Contributor·5 tasks·

  ·
  9

• [Performance] Add QLI candidate skip fast path (mode 4) for QuantLightningIndexerV2

  module:core
  #19911·qys4123 opened Oct 27, 2026·4 of 5 tasks·

  ·
  1

  7

• [Performance] Automatically budget FlashMLA DCP overlap cores

  module:opsmodule:tests
  #19909·pisceskkk opened Oct 27, 2026Contributor·4 of 5 tasks·

  ·
  6

• [Doc][Misc]Clarify Qwen3.8-2.4T-A95B deployment guide for v0.30.0

  documentation
  #19908·Jingluo-nan opened Oct 27, 2026·5 tasks·

  ·
  3

• [v0.30.0][Doc] Translated Doc files 2026-10-27

  documentation
  #19907·vllm-ascend-ci opened Oct 27, 2026Collaborator·

  ·
  2

• [Cherry-pick][releases/v0.31.0rc][CI] Select vLLM ref by PR base branch for E2E and CPU UT (from #19902)

  ci/builddocumentation
  #19906·vllm-ascend-ci opened Oct 27, 2026Collaborator·

  ·
  4