All pull requests

Pull requests

• test(routing): the no-hint spread test sends twelve distinct prompts, so a fair tie draw landing every miss on one worker is a one-in-177,147 run, not one in 243

  model-gatewaytests
  #4402·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  2

• feat(servicer): the Rust gRPC servicer is the default; the Python servicer is deprecated and a launch the Rust path does not serve falls back to it

  cidocumentationgrpcpython-bindingstests
  #4401·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  14

• fix(metrics): a collected answer records its token and generation families as a streamed one does

  grpcmodel-gateway
  #4399·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  4

• perf(tokenizer): an alternative with an optional prefix is entered only when the run after it can begin

  tokenizer
  #4395·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  5

• fix(kv_index): the string tree's snapshot merge compares a remote edge that continues below a local child against that child's own children

  kv-index
  #4394·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  14

• fix(gateway): the PD count-tokens route logs the worker and endpoint of a call that failed below HTTP

  model-gateway
  #4387·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  15

• feat(servicer): the Python servicer serves the Rust servicer's metrics families on --metrics-port

  documentationgrpctests
  #4385·hello-alexmcc opened Oct 24, 2026Collaborator·3 of 4 tasks·

  ·
  21

• fix(metrics): the registration step publishes a Symphony table's failure classes once the worker is in the registry

  grpcmodel-gatewaytests
  #4382·hello-alexmcc opened Oct 24, 2026Collaborator·

  ·
  7

• ci: record required vLLM unit results on AMD

  citests
  #4381·chunfangamd opened Oct 24, 2026Collaborator·4 tasks·

  ·
  2

• feat(observability): the media pipeline's stages, items and fetches are series on the router and the worker side

  grpcmodel-gatewaymultimodaltests
  #4377·hello-alexmcc opened Oct 23, 2026Collaborator·3 of 4 tasks·

  ·
  42

• fix(ci): preserve active model-download locks

  citests
  #4376·chunfangamd opened Oct 23, 2026Collaborator·4 tasks·

  ·
  2

• feat(metrics): a request's time in the pipeline is split by stage on smg_router_stage_seconds

  grpcmodel-gateway
  #4375·hello-alexmcc opened Oct 23, 2026Collaborator·

  ·
  22

• ci: retain AMD e2e JUnit reports and worker logs

  ci
  #4372·chunfangamd opened Oct 23, 2026Collaborator·

  ·
  3

• ci: check AMD e2e dependencies after install and cleanup

  citests
  #4371·chunfangamd opened Oct 23, 2026Collaborator·

  ·
  3

• test(discovery): the Consul e2e keeps the periodic pass, which applies a change that lands while its instance still registers

  model-gatewaytests
  #4370·hello-alexmcc opened Oct 23, 2026Collaborator·2 of 4 tasks·

  ·
  5

• fix(mcp): preserve engine HTTP requirements in CI

  cidependenciestests
  #4351·chunfangamd opened Oct 23, 2026Collaborator·

  ·
  2

• fix(grpc): generate stubs compatible with engine runtimes

  cidependenciesdocumentationgrpc
  #4350·chunfangamd opened Oct 23, 2026Collaborator·

  ·
  2

• feat(logging): the access-log record carries the generation's facts the pipeline knows when the response ends

  grpcmodel-gatewayprotocolspython-bindingstests
  #4347·hello-alexmcc opened Oct 23, 2026Collaborator·3 of 4 tasks·

  ·
  32

• feat(metrics): the router error counter carries the reason behind each error class

  grpcmodel-gateway
  #4345·hello-alexmcc opened Oct 23, 2026Collaborator·

  ·
  22

• ci: pair the October 23 TokenSpeed runtime with scheduler 0.1.42

  tests
  #4334·chunfangamd opened Oct 23, 2026Collaborator·

  ·
  4

• perf(tokenizer): the byte-level table knows at load which tokens are whole, so a step reads them without validation

  tokenizer
  #4323·hello-alexmcc opened Oct 23, 2026Collaborator·

  ·
  5

• ci: add AMD worker-side vLLM multimodal coverage

  ci
  #4298·chunfangamd opened Oct 23, 2026Collaborator·

  ·
  5

• fix(gateway): an explicit thinking.disabled wins over reasoning_effort

  grpcmodel-gatewayprotocols
  #4284·hello-alexmcc opened Oct 23, 2026Collaborator·

  ·
  6

• fix(servicer): the PD decode side withholds the KV-transfer rejection notice on the Mooncake connector

  grpctests
  #4281·hello-alexmcc opened Oct 23, 2026Collaborator·

  ·
  5

• fix(chat): the request's add_generation_prompt is read, so false renders the conversation without a generation header

  grpcmodel-gatewayprotocolsteststokenizer
  #4280·hello-alexmcc opened Oct 23, 2026Collaborator·

  ·
  19