All pull requests

Pull requests

• [None][test] Migrate model and executor fixtures to KV cache manager V2

  #20098·yizhang-nv opened Oct 11, 2026Member·1 task done·

  ·
  

• [None][test] Migrate attention fixtures and prims-ts validation to KV cache V2

  #20097·yizhang-nv opened Oct 11, 2026Member·1 task done·

  ·
  

• [None][test] Migrate encoder-decoder coverage to KV cache manager V2

  #20096·yizhang-nv opened Oct 11, 2026Member·1 task done·

  ·
  

• [None][feat] KV cache manager V2 for hybrid RL rollouts: batched core API, aligned snapshot offsets, folded save-last, refit with open caches

  #20095·zongfeijing opened Oct 11, 2026Collaborator·

  ·
  

• [None][fix] Declare fla-core for Kimi KDA runtime imports

  #20094·lorenzozanee opened Oct 11, 2026·

  ·
  1

  4

  

• [None][feat] Let a model require KV cache manager v2 instead of being demoted to v1

  #20093·brnguyen2 opened Oct 10, 2026Collaborator·1 task done·

  ·
  15

  

• [https://nvbugs/6901613][fix] Two-pass variance in the same kernel: reduce and share the mean, then sum…

  VisualGen
  #20092·trtllm-agent opened Oct 10, 2026Collaborator·2 tasks done·

  ·
  1

  

• [https://nvbugs/6890768][fix] Turn TF32 off (allow_tf32=False) around the stub's attention math with a…

  VisualGen
  #20091·trtllm-agent opened Oct 10, 2026Collaborator·2 tasks done·

  ·
  1

  

• [https://nvbugs/6891758][fix] Every lane calls fence_view_async_shared() and then __syncwarp() at the…

  #20090·trtllm-agent opened Oct 10, 2026Collaborator·2 tasks done·

  ·
  2

  

• [https://nvbugs/6891163][fix] Build embed_tokens as a full copy (no mapping) when enable_attention_dp and…

  ci: full pre-merge approved
  #20089·trtllm-agent opened Oct 10, 2026Collaborator·2 tasks done·

  ·
  2

  

• [https://nvbugs/6891165][fix] Pass chat_template_kwargs in the request body via extra_body; schema…

  ci: full pre-merge approved
  #20088·trtllm-agent opened Oct 10, 2026Collaborator·2 tasks done·

  ·
  2

  

• [None][feat] Add native KV cache V2 sleep and wakeup

  #20087·yizhang-nv opened Oct 10, 2026Member·1 task done·

  ·
  

• [None][feat] Add native KV cache retention to KVCM V2

  api-compatible
  #20086·yizhang-nv opened Oct 10, 2026Member·6 tasks done·

  ·
  

• [None][perf] Avoid cuBLAS's fallback in the KDA prefill projection

  #20084·ruocheng-nv opened Oct 10, 2026Contributor·1 task done·

  ·
  6

  

• [None][fix] Restore routing-weight dtype after NcclEP dispatch

  #20081·peaceh-nv opened Oct 10, 2026Collaborator·1 task done·

  ·
  1

  

• [None][infra] Retry build_wheel.py in the build stage

  #20080·weiminwang-nv opened Oct 10, 2026Contributor·4 tasks done·

  ·
  1

  

• [None][feat] Prototype faster native startup with Snapshot

  #20079·chienchunhung opened Oct 10, 2026Collaborator·1 of 8 tasks·

  ·
  3

  

• [TRTLLM-14620][feat] Integrate DSpark confidence scheduling and mixed verification

  #20078·zhaoyuanh-nvidia opened Oct 10, 2026Collaborator·1 task done·

  ·
  6

  

• [None][feat] Modeling V2 dynamic switch with LLM config and args

  api-compatible
  #20077·WeiHaocheng opened Oct 10, 2026Collaborator·1 task·

  ·
  14

  

• [None][test] Update MiniMax M3 EAGLE3 draft model directory

  #20076·fredricz-20070104 opened Oct 10, 2026Collaborator·

  ·
  1

  

• [None][feat] MXFP8 refit with PCG changes, recompute_active_requests, Ray collectives

  #20075·hchings opened Oct 10, 2026Collaborator·1 task·

  ·
  

• [TRTLLM-14620][feat] Add DSpark ragged graph and runtime integration

  #20072·zhaoyuanh-nvidia opened Oct 9, 2026Collaborator·1 task done·

  ·
  3

  

• [TRTLLM-14620][feat] Integrate calibrated DSpark confidence scoring

  #20071·zhaoyuanh-nvidia opened Oct 9, 2026Collaborator·1 task done·

  ·
  10

  

• [TRTLLMINF-336][infra] Re-enable BOLT premerge consume

  #20070·mlefeb01 opened Oct 9, 2026Collaborator·1 task done·

  ·
  4

  

• [None][feat] VC-Attention-QK16 recipe for PrimTS dense context FMHA

  VisualGen
  #20068·harrisonzhy opened Oct 9, 2026·1 task done·

  ·
  3