All pull requests

Pull requests

• test(sidecar): add isolated units beside production code

  actionsdocumentationsize/XXLtest
  #15243·JulienDarve opened Sep 24, 2026Contributor·

  ·

• build: compile NVTX ranges into runtime wheels via the header-only nvtx crate

  buildcontainersize/M
  #15242·peilii opened Sep 23, 2026Contributor·

  ·
  9

• fix(anthropic): open a fresh text block for text arriving after streamed tools

  external-contributionfixfrontendsize/Mtrusted-contributor
  #15241·michaelfeil opened Sep 23, 2026Contributor·

  ·
  7

• feat(replay): add conversation affinity to the existing adapter

  containerdocumentationfeatplannerroutersize/XXL
  #15240·Harrilee opened Sep 23, 2026Contributor·

  ·

• feat(mocker): support state-cache checkpoint reuse

  containerdocumentationfeatplannersize/XXL
  #15239·dreamtalen opened Sep 23, 2026Contributor·

  ·
  1

• fix(tests): arm the vLLM worker's startup VRAM guard in the frontend test

  external-contributionfixsize/M
  #15236·glamr-agent opened Sep 23, 2026Contributor·

  ·
  29

• perf(llm): share prompt tokens during decoder setup

  perfsize/S
  #15234·jthomson04 opened Sep 23, 2026Contributor·

  ·
  1

• fix(vllm): name the worker that ends a disaggregated multimodal launch

  backend::vllmexternal-contributionfixsize/L
  #15233·glamr-agent opened Sep 23, 2026Contributor·

  ·
  41

• Codex/router prefill cost only

  backend::vllmcontainersize/L
  #15232·NVShreyas opened Sep 23, 2026Contributor·

  ·
  7

• feat(runtime): add Velo TCP and UCX response transport

  documentationfeatfrontendsize/XXL
  #15231·jthomson04 opened Sep 23, 2026Contributor·

  ·

• fix(runtime): exit the ingress response pump when the request context is cancelled

  external-contributionfixsize/Ltrusted-contributor
  #15230·michaelfeil opened Sep 23, 2026Contributor·

  ·
  10

• fix(frontend): remove trailing DONE sentinel from Responses streams

  fixfrontendsize/L
  #15229·mattwittwer opened Sep 23, 2026·

  ·
  4

  

• fix(vllm): honor managed ports in the disagg multimodal E+PD launcher

  backend::vllmexternal-contributionfixsize/L
  #15228·glamr-agent opened Sep 23, 2026Contributor·

  ·
  65

• fix(sglang): gate auto tool-call grammars on reasoning

  external-contributionfixfrontendsize/M
  #15227·sherlockwu opened Sep 23, 2026Contributor·

  ·
  17

• fix(operator): reject vLLM parallelism values the engine would refuse

  deployment::k8sdocumentationfixsize/L
  #15226·tzulingk opened Sep 23, 2026Contributor·

  ·
  7

• fix(operator): derive the multinode topology from the full argv

  deployment::k8sdocumentationfixsize/M
  #15225·tzulingk opened Sep 23, 2026Contributor·

  ·
  2

• fix(operator): resolve repeated vLLM flags to their last occurrence

  deployment::k8sfixsize/M
  #15224·tzulingk opened Sep 23, 2026Contributor·

  ·
  3

• feat(power-agent): gate pod readiness on cap enforcement

  deployment::k8sfeatsize/XXL
  #15223·kaim-eng opened Sep 23, 2026Contributor·

  ·
  15

• docs(readme): add TensorRT-LLM quickstart examples

  docsdocumentationsize/S
  #15222·alec-flowers opened Sep 23, 2026Contributor·

  ·
  3

• feat(runtime): add frontend lifecycle state evidence

  external-contributionfeatfrontendroutersize/XXL
  #15221·sbak5 opened Sep 23, 2026Contributor·

  ·
  2

• fix(trtllm): Allow data: URI videos to be decoded with NVDEC

  backend::trtllmexternal-contributionfixmultimodalsize/L
  #15219·aswinvisva opened Sep 23, 2026Contributor·

  ·
  37

• feat(recipes): add Motif-3 NVFP4 serving recipe.

  documentationexternal-contributionfeatsize/L
  #15217·fsaady opened Sep 23, 2026Contributor·

  ·
  2

• fix(operator): support B300 in DGDR GPU discovery

  deployment::k8sdocumentationfixplannersize/M
  #15216·sttts opened Sep 23, 2026Contributor·

  ·
  15

• chore: bump trtllm to v1.3.0rc28

  backend::trtllmchorecontainerdep-upgradedocumentationsize/XXL
  #15215·dynamo-ops opened Sep 23, 2026Contributor·

  ·
  1

• fix(benchmarks): verify router worker participation

  documentationexternal-contributionfixsize/XL
  #15213·cpakkamisaac-sae opened Sep 23, 2026Contributor·

  ·
  1

  28