All pull requests

Pull requests

• fix(backend-common): fetch HF models without weights at Worker registration

  fixsize/M
  #18295·jthomson04 opened Nov 6, 2026Contributor·1 task done·

  ·
  1

  1

• fix(sglang): import MultimodalProcessorOutput from req_types

  backend::sglangexternal-contributionfixsize/XS
  #18293·YAMY1234 opened Nov 6, 2026Contributor·1 task·

  ·
  5

• fix(benchmarks): fail closed on missing or invalid aggregate exports

  documentationfixsize/XL
  #18292·ajcasagrande opened Nov 6, 2026Contributor·1 task done·

  ·
  1

• docs(sglang): fix memory-management engine route paths

  docsdocumentationexternal-contributionsize/M
  #18290·glamr-agent opened Nov 6, 2026Collaborator·

  ·
  27

• perf(sglang,trtllm): make PNG compression configurable for image workers

  backend::sglangbackend::trtllmdocumentationexternal-contributionsize/L
  #18289·glamr-agent opened Nov 6, 2026Collaborator·

  ·
  11

• fix(gms): accept the vLLM 0.32+ request_memory signature in shadow mode

  external-contributionfixsize/M
  #18288·glamr-agent opened Nov 6, 2026Collaborator·

  ·
  13

• test(replay): [cherry-pick 1.7.0] detect AISimulate under the aisim distribution name (#18286)

  size/Stest
  #18287·BenHamm opened Nov 6, 2026Contributor·

  ·

• test(replay): detect AISimulate under the aisim distribution name

  size/Stest
  #18286·BenHamm opened Nov 6, 2026Contributor·

  ·
  3

• fix(vllm): preserve native decoder KV coordinates during prefill

  backend::vllmdocumentationfixsize/L
  #18285·yifjiang opened Nov 6, 2026Contributor·1 task done·

  ·
  1

• fix(multimodal): validate audio and image batches before creating load coroutines

  external-contributionfixmultimodalsize/L
  #18284·glamr-agent opened Nov 6, 2026Collaborator·

  ·
  6

• docs(snapshot): use nvidia.com/snapshot-multi-gpu for the opt-out annotation

  docsdocumentationexternal-contributionsize/XS
  #18283·glamr-agent opened Nov 6, 2026Collaborator·1 task·

  ·
  10

• docs(observability): document NVTX profiling and DYN_NVTX

  docsdocumentationsize/L
  #18282·nv-kmcgill53 opened Nov 6, 2026Contributor·1 task·

  ·
  12

• refactor(router): write worker-selection policies inline per role

  documentationexternal-contributionrefactorsize/XXL
  #18281·ishandhanani opened Nov 6, 2026Contributor·

  ·
  6

• ci: force-update the docs-website ref in the composition check

  actionscisize/XS
  #18279·dagil-nvidia opened Nov 6, 2026Collaborator·1 task·

  ·
  3

• fix(recipes): update Gemma4 TensorRT-LLM configuration

  fixsize/XS
  #18275·2ez4bz opened Nov 5, 2026Contributor·1 task·

  ·
  2

• feat(dynamo): Allow OpenTelemetry logs/traces to be enabled independently

  documentationexternal-contributionfeatsize/L
  #18274·CanvaMatt opened Nov 5, 2026·1 task done·

  ·
  2

• fix(examples): declare decode mode for Global Planner XPU decode worker

  external-contributionfixsize/XSxpu
  #18273·glamr-agent opened Nov 5, 2026Collaborator·

  ·
  8

• test(sglang): simulate the Rust server's stream shape in the format laws

  size/Stest
  #18270·jain-ria opened Nov 5, 2026Contributor·1 task·

  ·
  2

• test(router): add overload protection e2e mocker test

  documentationsize/XLtest
  #18269·kthui opened Nov 5, 2026Contributor·1 task done·

  ·
  5

• fix(sglang-sidecar): reject out-of-vocabulary native input token IDs

  external-contributionfixsize/L
  #18268·glamr-agent opened Nov 5, 2026Collaborator·

  ·
  7

• fix(trtllm): abort cancelled requests when stream cleanup cancels the monitor

  backend::trtllmexternal-contributionfixsize/L
  #18265·glamr-agent opened Nov 5, 2026Collaborator·

  ·
  10

• fix(frontend): preserve GPT-OSS return-token tool calls

  fixfrontendsize/L
  #18263·rmccorm4 opened Nov 5, 2026Contributor·1 task·

  ·
  1

• fix(sglang): decode NVDEC video frames into device memory

  backend::sglangexternal-contributionfixmultimodalsize/M
  #18262·glamr-agent opened Nov 5, 2026Collaborator·1 task done·

  ·
  15

• perf(frontend): write common chat SSE chunks without serde's map serializer

  frontendperfsize/L
  #18260·jthomson04 opened Nov 5, 2026Contributor·1 task·

  ·
  1

• feat(recipes): add Qwen3.8-27B agent recipe on the vLLM sidecar

  documentationfeatsize/L
  #18259·KrishnanPrash opened Nov 5, 2026Contributor·1 task done·

  ·
  6