All pull requests

Pull requests

• [https://nvbugs/7029684][fix] Precompile MiniMax FMHA variants in wheels

  #21443·ruodil opened Nov 2, 2026Collaborator·1 task·

  ·
  

• [TRTLLM-17262][refactor] Split the disaggregated server's entry-point wrapper into prepare, run, finish and serialize steps

  #21442·JunyiXu-nv opened Nov 2, 2026Collaborator·1 task done·

  ·
  

• [TRTLLM-17557][fix] Make the configuration-database CLI test runnable

  #21441·JunyiXu-nv opened Nov 2, 2026Collaborator·1 task done·

  ·
  3

  

• [TRTLLM-17524][feat] Switch to lld linker

  #21440·Funatiq opened Nov 2, 2026Collaborator·1 task·

  ·
  3

  

• [None][test] Refresh B200 VisualGen perf sanity recipes from the October sweep

  VisualGen
  #21439·taianz-nv opened Nov 2, 2026Collaborator·

  ·
  

• [None][feat] DeepSeek-V4.1: Perf optimizations for DS-V4.1-Flash-NVFP4

  #21438·ziyixiong-nv opened Nov 2, 2026Collaborator·1 task·

  ·
  2

  

• [https://nvbugs/7029762][docs] Link current PyTorch KV cache managers

  #21437·ruodil opened Nov 2, 2026Collaborator·1 task done·

  ·
  4

  

• [https://nvbugs/7029761][docs] Qualify dynamic-tree sliding-window compatibility

  #21436·ruodil opened Nov 2, 2026Collaborator·1 task done·

  ·
  4

  

• [https://nvbugs/7029760][docs] Split parallelism YAML into independent examples

  #21435·ruodil opened Nov 2, 2026Collaborator·1 task done·

  ·
  4

  

• [#20969][fix] Qwen3.5/3.8: default dense NVFP4 checkpoints to Marlin on Ada/Hopper

  #21434·Smallfu666 opened Nov 2, 2026·1 task done·

  ·
  1

  1

  

• [None][feat] Add a cached-context multi-turn mode to benchmark_serving

  #21433·dominicshanshan opened Nov 2, 2026Collaborator·1 task done·

  ·
  3

  

• [TRTLLM-17505][feat] Add the SINK_REQUIRED_ONLY attention reuse policy to KV Cache Manager v2

  #21432·Hudayday opened Nov 2, 2026Collaborator·

  ·
  1

  

• [TRTLLMINF-577][infra] increase Python dependency installation timeout to 60 minutes

  ci: full pre-merge approved
  #21430·hanjingtian opened Nov 2, 2026Collaborator·1 task done·

  ·
  5

  

• [None][test] Add Rubin (GR100) single-GPU perf-sanity cases

  #21429·chenfeiz0326 opened Nov 2, 2026Collaborator·1 task done·

  ·
  4

  

• [TRTLLM-17519][refactor] Merge PoolingRunner into EncoderRunner with packed and scheduled entry points

  #21428·lori-ren opened Nov 2, 2026Contributor·1 task done·

  ·
  

• [TRTLLMINF-576][infra] Regenerate results-timeout.xml from the current unfinished_test.txt

  #21427·yiqingy0 opened Nov 2, 2026Collaborator·1 task·

  ·
  3

  

• [None][test] Add Muse Spark Open NVFP4 core performance coverage

  #21426·yufeiwu-nv opened Nov 2, 2026Collaborator·

  ·
  6

  

• [None][perf] strict_accept: one compiled variant for every batch

  ci: full pre-merge approved
  #21424·xguannv opened Nov 2, 2026Contributor·1 task done·

  ·
  4

  

• [None][perf] CuTe DSL MLA: admit FP8 wide-head verify groups on SM107

  #21423·xguannv opened Nov 2, 2026Contributor·1 task done·

  ·
  4

  

• [None][feat] add nemotron-h bf16 megamoe

  #21422·nv-guomingz opened Nov 2, 2026Collaborator·1 task·

  ·
  

• [None][test] extend Qwen3 NVFP4 CUTLASS RTX test timeout

  ci: full pre-merge approved
  #21421·jieli-matrix opened Nov 2, 2026Collaborator·1 task done·

  ·
  9

  

• [None][doc] Update MoE perf guide and dashboard with VR Rubin

  #21420·leslie-fang25 opened Nov 2, 2026Collaborator·1 task done·

  ·
  4

  

• [None][perf] autotuner: share one CUDA graph pool across a tuning pas…

  #21419·xguannv opened Nov 2, 2026Contributor·1 task done·

  ·
  1

  

• [None][infra] select covered Python files across repository directories

  #21418·crazydemo opened Nov 2, 2026Collaborator·1 task done·

  ·
  14

  

• [None][fix] Wait for the rank-0 VisualGen worker teardown at external-launch shutdown

  ci: full pre-merge approvedVisualGen
  #21417·luyiyun1021 opened Nov 2, 2026Collaborator·1 task done·

  ·
  16