All pull requests

Pull requests

• [#22152][fix] Notify batched executor responses on each request event loop

  #22153·itsrayeedscodespace opened Nov 11, 2026·1 task done·

  ·
  1

  4

  

• [#22150][fix] Register RPC response futures before sending requests

  #22151·itsrayeedscodespace opened Nov 11, 2026·1 task done·

  ·
  1

  2

  

• [None][perf] Muse Spark GB300: fmk context FMHA for mixed steps on sm_103 (opt-in)

  #22149·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  1

  

• [None][perf] Muse Spark GB300: one-cluster sliding-window attention block (CL8) for one-request decode

  #22148·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  1

  

• [None][perf] Muse Spark GB300: router GEMM 8-aligned row tiles (opt-in)

  #22147·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  1

  

• [None][perf] Muse Spark GB300: persistent-expert tactic override and graph-launch scheduling reset (opt-in)

  #22146·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  1

  

• [None][perf] Muse Spark GB300 TEP4: opt-in extra fast paths of the TEP8 composition

  #22145·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  2

  

• [None][perf] KVCacheV2Scheduler: prepare context caches ahead on interval-deferred passes

  #22144·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  2

  

• [None][perf] Muse Spark GB300: split-KV decode reads 64-token KV cache blocks

  #22143·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  2

  

• [None][perf] Muse Spark GB300: padding-slot one-hot draft probs without a host sync

  #22142·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  2

  

• [None][feat] Muse Spark GB300: block-verified draft sampling (#21475 cherry-pick) with fused verify launches

  #22141·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  1

  

• [None][perf] Muse Spark GB300: configurable trtllm-serve HTTP keep-alive timeout

  #22140·viraatc opened Nov 11, 2026·3 tasks done·

  ·
  2

  

• [None][fix] Count reclaimable page cache as available in the rank-striped memory check

  #22139·moraxu opened Nov 11, 2026Collaborator·1 task done·

  ·
  

• [None][test] Stop forcing the HF weight cache in test session reuse

  #22138·moraxu opened Nov 11, 2026Collaborator·1 task done·

  ·
  

• [None][infra] Add a report for the checkpoint I/O CI startup experiment

  #22137·moraxu opened Nov 11, 2026Collaborator·1 task done·

  ·
  

• [TRTLLM-14886][feat] Qualify DeepSeek-V3 (MLA + CUTLASS MoE) for MX

  #22136·moraxu opened Nov 11, 2026Collaborator·1 task·

  ·
  

• [TRTLLM-14893][feat] Qualify the Qwen3.5 MoE vision-language root (CUTLASS BF16) for MX

  #22135·moraxu opened Nov 11, 2026Collaborator·4 of 5 tasks·

  ·
  

• [#22133][fix] Enqueue a multi-input embedding request all or nothing

  Community want to contribute
  #22134·dundysm opened Nov 11, 2026·1 task done·

  ·
  1

  3

  

• [#22131][fix] Keep a re-added server's load when its old requests finish

  Community want to contribute
  #22132·dundysm opened Nov 11, 2026·1 task done·

  ·
  1

  9

  

• [https://nvbugs/7082289][infra] Waive 1 failed case(s) on CPU

  ci: full pre-merge approved
  #22129·trtllm-agent opened Nov 11, 2026Collaborator·1 task·

  ·
  12

  

• [#22125][fix] Name max_batch_size in the V2 Mamba GPU quota error

  Community want to contributestalewaiting for feedback
  #22127·dundysm opened Nov 11, 2026·3 of 4 tasks·

  ·
  1

  7

  

• [None][fix] Build the head-104 FMHA regression from components

  #22124·chzblych opened Nov 11, 2026Collaborator·3 tasks done·

  ·
  1

  

• [None][feat] Add a modeling-task-yaml skill to help modeling v2 onboard

  #22123·WeiHaocheng opened Nov 11, 2026Collaborator·1 task done·

  ·
  12

  

• [None][refactor] Move Mamba2/GDN speculative replay state into per-family providers

  #22121·VALLIS-NERIA opened Nov 11, 2026Collaborator·

  ·
  6

  

• [None][feat] Add VisualGen layer and step quantization exclusions

  api-compatibleVisualGen
  #22120·yumin066 opened Nov 11, 2026Collaborator·1 task·

  ·