All pull requests

Pull requests

• fix: stop /v1/responses asking for logprobs on every request

  #19665·v0ropaev opened Sep 28, 2026·

  ·
  1

  

• [#19659][fix] Bound remote MPI failures and launcher shutdown

  #19664·chienchunhung opened Sep 28, 2026Collaborator·

  ·
  

• [None][test] Add KV transfer lifecycle acceptance regressions

  #19663·chienchunhung opened Sep 28, 2026Collaborator·

  ·
  

• [https://nvbugs/6763479][chore] Update GLM-Image goldens

  VisualGen
  #19661·jloftin-nv opened Sep 28, 2026Collaborator·

  ·
  7

  

• [None][fix] Fail fast in trtllm-llmapi-launch when the rank-0 task exits and the worker world is wedged

  #19660·whiz-Tuhin opened Sep 28, 2026·

  ·
  1

  10

  

• [None][perf] Accept a flat int32 array as GenerationRequest prompt without materializing a list

  #19658·peilii opened Sep 28, 2026·

  ·
  7

  

• [TRTLLM-16670][feat] Add MiniMax H3 ref2va mode support

  VisualGen
  #19657·yibinl-nvidia opened Sep 28, 2026Collaborator·

  ·
  

• [https://nvbugs/6831694][fix] Unwaive MSA cache-view and warmup policy tests

  #19655·peihu-nv opened Sep 28, 2026Collaborator·

  ·
  4

  

• [None][fix] Gate Flash-Next FP8 concurrency-16 perf on B300/GB300

  #19654·yufeiwu-nv opened Sep 28, 2026Collaborator·

  ·
  4

  

• [None][perf] Fill the KVCM2 host tier progressively

  #19650·lowsfer opened Sep 28, 2026Member·

  ·
  26

  

• [TRTLLM-15824][feat] Enable EVS for Nemotron Super 3.5 VL

  #19648·eopXD opened Sep 28, 2026Collaborator·

  ·
  7

  

• [None][feat] Add DynamicEPLB kernels and runtime integration

  api-compatible
  #19646·JintaoPengCS opened Sep 28, 2026Collaborator·

  ·
  17

  

• [TRTLLM-16599][perf] Fuse MiniMax-H3 BF16 rotary embeddings

  VisualGen
  #19645·karljang opened Sep 27, 2026Collaborator·

  ·
  1

  

• [None][fix] Move the CuteDSL locality domain weight split into staged hooks

  #19644·zhangcl opened Sep 27, 2026Collaborator·

  ·
  1

  

• [https://nvbugs/6697099][fix] Keep MPI progress alive on the PP sample-state relay thread

  #19643·YihuiLu512 opened Sep 27, 2026Collaborator·

  ·
  7

  

• [None][test] Remove 32 closed-bug waive entries for main

  #19642·trtllm-agent opened Sep 27, 2026Collaborator·

  ·
  

• [None][fix] Pass max_seq_len to the Qwen-VL attention helper in the MiniCPM-V 4.6 vision encoder

  Community want to contribute
  #19641·David-Wu1119 opened Sep 26, 2026Contributor·

  ·
  2

  

• [None][fix] add FP16 support to fused QK norm RoPE

  Community want to contribute
  #19640·atipre opened Sep 26, 2026·

  ·
  3

  

• [None][test] Skip MoE A2A CFT regression under CUDA forward compatibility

  #19639·ruocheng-nv opened Sep 26, 2026Contributor·

  ·
  6

  

• [#19635][fix] Reject guided decoding without a backend

  Community want to contribute
  #19638·lorenzozanee opened Sep 26, 2026·

  ·
  1

  1

  

• [https://nvbugs/6616036][test] Unwaive test

  #19637·dominicshanshan opened Sep 26, 2026Collaborator·

  ·
  4

  

• [https://nvbugs/6801102][fix] Registered the measured GSM8K reference (89.083) for MXFP8 + FP8 KV cache…

  #19633·trtllm-agent opened Sep 25, 2026Collaborator·

  ·
  4

  

• [https://nvbugs/6831694][fix] Change the stale expected-call literal to assert_called_once_with(3) (still…

  #19631·trtllm-agent opened Sep 25, 2026Collaborator·

  ·
  1

  

• [https://nvbugs/6831653][fix] Test-only one-line fix: the assertion becomes assert_called_once_with(3)…

  #19630·trtllm-agent opened Sep 25, 2026Collaborator·

  ·
  1

  

• [None][fix] perf-optimize: retry a stage that returned without its deliverable

  #19629·hyukn opened Sep 24, 2026Collaborator·

  ·