All pull requests

Pull requests

• [TRTLLM-17265][refactor] Split openai_protocol into a package

  VisualGen
  #21544·JunyiXu-nv opened Nov 3, 2026Collaborator·1 task done·

  ·
  

• [TRTLLM-17612][refactor] Remove unused kernels after TensorRT backend removal

  #21543·Funatiq opened Nov 3, 2026Collaborator·1 task·

  ·
  2

  

• [None][test] State the GPU count of multi-GPU QA core tests with markers

  #21541·jieli-matrix opened Nov 3, 2026Collaborator·1 task done·

  ·
  

• [None][feat] Run MiniMax-M3.1-Flash on Hopper

  #21540·ischencheng opened Nov 3, 2026·1 task done·

  ·
  3

  

• [None][fix] Report the full output width in the INT4 GEMM fakes

  #21539·0z5a opened Nov 3, 2026Contributor·1 task done·

  ·
  5

  

• [None][fix] Port native disaggregation KV result report token - #20723

  #21538·BoyueZ opened Nov 3, 2026Collaborator·1 task·

  ·
  

• [None][feat] Add Heun sampler to VisualGen, enabled for MiniMax-H3

  VisualGen
  #21537·jhaotingc opened Nov 3, 2026Collaborator·1 task done·

  ·
  1

  

• [None][doc] Release notes for 1.3 release

  #21536·BowenFu opened Nov 3, 2026Contributor·4 tasks done·

  ·
  12

  

• [None][feat] DFlash: TRTLLM drafter attention by default wherever it runs, cold-page compression with a TRTLLM drafter, and acceptance-length tests

  #21535·zheyuf opened Nov 3, 2026Collaborator·1 task·

  ·
  

• [TRTLLMINF-586][infra] Add platform context to LLMAPI test stages

  ci: full pre-merge approved
  #21534·mzweilz opened Nov 3, 2026Collaborator·1 task done·

  ·
  4

  

• [None][fix] Disaggregated generation: one-hot draft probabilities for placeholder drafts at the handoff

  #21533·viraatc opened Nov 3, 2026·2 tasks done·

  ·
  

• [None][perf] sampling: keep more loads in flight for small speculative batches

  #21532·Wanli-Jiang opened Nov 3, 2026Collaborator·1 task·

  ·
  3

  

• [None][fix] disaggregated serving: finish single-token generation requests with the first token

  #21531·Wanli-Jiang opened Nov 3, 2026Collaborator·1 task·

  ·
  6

  

• [None][fix] Port #17850 to feat/k3_26.08: keep the KV cache estimation managers off the host tier

  #21530·viraatc opened Nov 3, 2026·2 tasks done·

  ·
  

• [None][perf] Kimi K3: split-K FP8-KV MLA context FMHA for SM100/SM103

  #21529·viraatc opened Nov 3, 2026·3 tasks done·

  ·
  

• [https://nvbugs/6308608][fix] Apply one local media input policy to chat and VisualGen and rename its variables

  VisualGen
  #21528·BoyueZ opened Nov 3, 2026Collaborator·1 task done·

  ·
  4

  

• [TRTLLM-17123][feat] Gather cached MLA KV for Helix prefill

  #21527·zhaoyuanh-nvidia opened Nov 3, 2026Collaborator·1 task done·

  ·
  

• [TRTLLM-17122][fix] Sync owner-only MLA prefill review fixes to Helix aggregate

  #21526·zhaoyuanh-nvidia opened Nov 3, 2026Collaborator·1 task done·

  ·
  

• [None][chore] Deprecate KV cache manager V1 and the C++ cache transceiver

  api-compatible
  #21525·Shixiaowei02 opened Nov 3, 2026Collaborator·1 task done·

  ·
  8

  

• [None][infra] Retire the legacy L0 perf stage and dedicated node selection

  #21524·chzblych opened Nov 3, 2026Collaborator·1 task done·

  ·
  4

  

• [None][feat] Simplify per-view KV head transfer

  #21523·chuangz0 opened Nov 3, 2026Collaborator·1 task done·

  ·
  

• [None][fix] Match the fp8_quantize_1x128 fake to the real op on SM120

  #21522·0z5a opened Nov 3, 2026Contributor·1 task done·

  ·
  3

  

• [#21519][fix] Count decoding iterations across benchmark turns

  #21520·codebreaker77 opened Nov 3, 2026·1 task done·

  ·
  1

  1

  

• [TRTLLM-17263][refactor] Move the Anthropic serving modules into serve/api/anthropic

  #21518·JunyiXu-nv opened Nov 3, 2026Collaborator·

  ·
  

• [https://nvbugs/6428124][test] Unwaive test_lora_pp2_sanity

  #21517·BowenFu opened Nov 3, 2026Contributor·2 tasks done·

  ·
  4