All pull requests

Pull requests

• [None][perf] Enable FlashInfer GDN chunked prefill on SM107

  #21210·aswinvisva opened Oct 28, 2026Collaborator·1 task done·

  ·
  2

  

• [None][perf] Fuse residual add, RMSNorm and static FP8 quantization for the FP8 consumers of Qwen3-Next norms

  #21209·aswinvisva opened Oct 28, 2026Collaborator·1 task done·

  ·
  1

  

• [None][perf] Quantize large bf16 activations of static-FP8 linears with a Triton kernel

  #21208·aswinvisva opened Oct 28, 2026Collaborator·1 task done·

  ·
  2

  

• [None][perf] Fuse the GDN prefill causal conv1d with the post-conv split, norm and gating

  #21207·aswinvisva opened Oct 28, 2026Collaborator·1 task done·

  ·
  2

  

• [None][perf] Use larger multi-row tiles for prefill-sized gated RMSNorm launches

  #21206·aswinvisva opened Oct 28, 2026Collaborator·1 task done·

  ·
  2

  

• [https://nvbugs/6988813][test] Unwaive remote MPI submit_sync

  #21205·chienchunhung opened Oct 28, 2026Collaborator·1 task done·

  ·
  2

  

• [https://nvbugs/7001101][fix] Put the coordinator Unix socket in an owner-only directory

  #21204·trtllm-agent opened Oct 28, 2026Collaborator·1 task done·

  ·
  5

  

• [https://nvbugs/6982378][fix] Build the FlashInfer FMHA module on all PP ranks before warmup

  #21203·trtllm-agent opened Oct 28, 2026Collaborator·1 task done·

  ·
  4

  

• [None][infra] Add skasetty-Siri to blossom-ci authorized users

  #21202·ishovkun opened Oct 28, 2026Contributor·1 task done·

  ·
  1

  

• [TRTLLM-17467][chore] Remove phantom dspark_confidence import hook from accept_stats

  #21201·brnguyen2 opened Oct 28, 2026Collaborator·3 tasks done·

  ·
  8

  

• [None][refactor] Adopt per-family native libraries for K3 modeling v2

  #21200·litaotju opened Oct 28, 2026Collaborator·1 task done·

  ·
  41

  

• [https://nvbugs/6858505][fix] Reject an encoder input longer than one encoder iteration at admission

  #21199·trtllm-agent opened Oct 28, 2026Collaborator·

  ·
  3

  

• [None][perf] Add opt-in tiled QKV preprocessing

  #21198·2ez4bz opened Oct 28, 2026Collaborator·1 task done·

  ·
  7

  

• [None][fix] Correct TPOT metric direction in perf sanity

  ci: full pre-merge approved
  #21197·dhansen-nvidia opened Oct 28, 2026Collaborator·

  ·
  8

  

• [None][feat] Close the Mooncake store feature gap: async loads, attention DP, speculative prefill and KV connector support in KVCacheManagerV2

  #21195·SimengLiu-nv opened Oct 28, 2026Collaborator·

  ·
  

• [None][feat] Attach feature and sparse metadata state as extensions

  VisualGen
  #21194·yihwang-nv opened Oct 28, 2026Collaborator·1 task·

  ·
  

• [https://nvbugs/6858547][fix] Ignore cp_type with one CP rank and require HELIX with more

  #21192·trtllm-agent opened Oct 28, 2026Collaborator·3 tasks done·

  ·
  1

  

• [TRTLLM-17105][fix] Report a bounce failure only after the receiver has answered every sent chunk

  #21191·chuangz0 opened Oct 28, 2026Collaborator·1 task done·

  ·
  4

  

• [None][fix] CuteDSL MoE: keep inplace FC1 outputs out of the Rubin autotune key

  #21190·Barry-Delaney opened Oct 28, 2026Collaborator·1 task done·

  ·
  1

  

• [TRTLLMINF-572][infra] Support setting separate rerun policies for each stages

  #21189·yiqingy0 opened Oct 28, 2026Collaborator·1 task·

  ·
  3

  

• [TRTLLMINF-170][infra] apply scoped CI node placement policies

  #21188·hanjingtian opened Oct 28, 2026Collaborator·1 task done·

  ·
  

• [https://nvbugs/6813945][fix] Use stored metadata server URLs as is and support etcd credentials and signed server records

  #21187·BoyueZ opened Oct 28, 2026Collaborator·1 task·

  ·
  

• [TRTLLM-17450][feat] Map KV cache pages between ranks from their page layouts

  #21186·bo-nv opened Oct 28, 2026Collaborator·1 task·

  ·
  

• [TRTLLM-14024][feat] Prune CuTe DSL GEMM autotuner tactics with nvMatmulHeuristics

  api-compatible
  #21185·peaceh-nv opened Oct 28, 2026Collaborator·1 task done·

  ·
  18

  

• [TRTLLM-17449][feat] Let KVCacheManagerV2 build its page table from layout hooks

  #21184·bo-nv opened Oct 28, 2026Collaborator·1 task·

  ·
  1