All pull requests

Pull requests

• [TRTLLM-15764][refactor] Split PyTorchModelEngine construction into named phases

  #20851·lori-ren opened Oct 23, 2026Contributor·1 task done·

  ·
  

• [https://nvbugs/6504651][fix] Add optional native RankInfo authentication

  #20850·yibinl-nvidia opened Oct 23, 2026Collaborator·

  ·
  

• [None][fix] purge unwanted CA certs

  #20849·tburt-nv opened Oct 23, 2026Collaborator·1 task·

  ·
  2

  

• [None][perf] MiniMax-M3 sparse attention: trim per-step host work

  #20848·pranav-nvidia opened Oct 23, 2026Contributor·1 task done·

  ·
  

• [None][feat] MiniMax-M3: load the compressed-tensors NVFP4 MoE packaging

  #20847·pranav-nvidia opened Oct 23, 2026Contributor·1 task done·

  ·
  

• [None][perf] GVR V2 top-K: in-kernel live-length admission for frozen CUDA-graph plans

  #20846·longcheng-nv opened Oct 23, 2026Collaborator·1 task done·

  ·
  5

  

• [https://nvbugspro.nvidia.com/bug/6820152][fix] Remove Wan I2V example test requiring unavailable NVFP4 checkpoint

  VisualGen
  #20845·o-stoner opened Oct 22, 2026Collaborator·1 task done·

  ·
  3

  

• [None][feat] Muse Spark Open modeling_v2 TEP target: mid-high-concurrency optimizations for Vera Rubin

  #20844·xwang233 opened Oct 22, 2026Collaborator·1 task done·

  ·
  1

  

• [None][chore] Mass merge release/1.3 branch back to main (batch 2)

  #20843·SimengLiu-nv opened Oct 22, 2026Collaborator·1 task·

  ·
  8

  

• [TRTLLM-17241][fix] Start trtllm-render on a host with no GPU driver

  #20842·hchings opened Oct 22, 2026Collaborator·1 task done·

  ·
  8

  

• [None][feat] Support shared-prefix attention for n-sample requests in FMHA

  #20841·hchings opened Oct 22, 2026Collaborator·1 task·

  ·
  

• [https://nvbugs/6939213][fix] Release cached MoE runner workspaces at executor shutdown

  #20840·trtllm-agent opened Oct 22, 2026Collaborator·

  ·
  1

  

• [https://nvbugs/6971880][fix] Sync serving schemas on version bumps

  #20839·Mgluhovskoi opened Oct 22, 2026Collaborator·1 task done·

  ·
  8

  

• [#20376][fix] Register grouped MLA generation cubins as Causal

  #20838·saiyambharara opened Oct 22, 2026·

  ·
  1

  2

  

• [None][fix] MiniMax-M3: resolve an unprefilled prompt as outside reasoning

  #20837·pranav-nvidia opened Oct 22, 2026Contributor·1 task done·

  ·
  8

  

• [None][perf] Wan 2.2: FP8 attention routing and static-FP8 block fusions on Rubin

  VisualGen
  #20836·wu6u3tw opened Oct 22, 2026Contributor·

  ·
  2

  

• [None][fix] use tempfile

  #20835·tburt-nv opened Oct 22, 2026Collaborator·1 task done·

  ·
  7

  

• [TRTLLM-17121][feat] Support phase-dependent MLA heads for Helix prefill

  #20833·zhaoyuanh-nvidia opened Oct 22, 2026Collaborator·1 task done·

  ·
  6

  

• [None][build] make GoogleTest discovery prefixes optional

  #20832·Funatiq opened Oct 22, 2026Collaborator·1 task done·

  ·
  6

  

• [#17022][fix] Honor --preserve_caller_max_tokens on trtllm-eval text tasks

  #20831·zw-g opened Oct 22, 2026·1 task done·

  ·
  1

  4

  

• [TRTLLM-16810][perf] SID-GR 3/N: Reduce per-step host overhead for wide beam search

  ci: full pre-merge approved
  #20830·mikeiovine opened Oct 22, 2026Collaborator·1 task done·

  ·
  7

  

• [None][infra] Select modeling_v2 CI tests by the node a change lands in

  #20829·WeiHaocheng opened Oct 22, 2026Collaborator·4 tasks done·

  ·
  15

  

• [https://nvbugs/6949270][fix] Size the Wan2.2-I2V serve test start budget to a cold checkpoint read

  VisualGen
  #20827·trtllm-agent opened Oct 22, 2026Collaborator·

  ·
  1

  

• [https://nvbugs/6739116][fix] Account for eager output memory when sizing KV cache

  #20826·VALLIS-NERIA opened Oct 22, 2026Collaborator·

  ·
  1

  

• [None][perf] MiniMax-M3: select context-row top-16 blocks with a CuTe DSL kernel

  #20825·zheyuf opened Oct 22, 2026Collaborator·1 task done·

  ·