All pull requests

Pull requests

• [None][fix] DSpark: keep the drafter's trained mask-token embedding

  Community want to contribute
  #19842·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [None][feat] Kimi K3 on modeling_v2: stateful catalog entries, MNNVL and collective decode kernels, drafter attention, MoE route B

  Community want to contribute
  #19841·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [https://nvbugs/6852473][fix] Skip zero-row launches in fp4_quantize, mxfp8_quantize and noaux_tc_op

  Community want to contribute
  #19840·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [None][feat] modeling_v2: Kimi K3 MXFP4 target, sm_100, tp16 attention + moe tp4 x ep4

  Community want to contribute
  #19839·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [#19819][fix] Prevent zero-sized patch grid and ZeroDivisionError in Nemotron dynamic resolution tiler

  Community want to contribute
  #19838·Ritabanm opened Oct 3, 2026·1 task done·

  ·
  1

  7

  

• [None][fix] Kimi Linear / Kimi K3: make fla-core a runtime dependency

  Community want to contribute
  #19837·vsabavat opened Oct 3, 2026·7 of 10 tasks·

  ·
  

• [None][fix] MNNVL all-reduce: cluster barrier use in the RMSNorm-fused kernels

  Community want to contribute
  #19836·vsabavat opened Oct 3, 2026·7 of 10 tasks·

  ·
  

• [None][fix] kda_prefill: return a new output tensor on every call

  Community want to contribute
  #19835·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [None][feat] modeling_v2 catalog: Kimi K3's generic-path entries and sm_100 cells

  Community want to contribute
  #19834·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [None][feat] modeling_v2 catalog: Kimi K3 single-GPU decode kernels as entries (sm_100)

  Community want to contribute
  #19833·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  

• [None][feat] modeling_v2 catalog: Kimi K3 KDA and MLA decode kernels as stateful entries (sm_100)

  Community want to contribute
  #19832·vsabavat opened Oct 3, 2026·8 of 10 tasks·

  ·
  16

  

• [None][fix] multi_stream_utils: record the calling stream on fn1's outputs

  Community want to contribute
  #19831·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  4

  

• [None][fix] kdaDecode legacy: __syncwarp before the lane-0 block-reduce stores

  Community want to contribute
  #19830·vsabavat opened Oct 2, 2026·7 of 10 tasks·

  ·
  1

  

• [None][fix] Kimi K3 attn_res kernels: order the ws_stats reads and fence the slot releases

  Community want to contribute
  #19829·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  2

  

• [None][fix] MNNVL all-reduce: keep a grown workspace's predecessor alive for captured graphs

  Community want to contribute
  #19828·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  6

  

• [None][fix] DFlash: start the dummy slot at context length 0 every step

  Community want to contribute
  #19827·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  1

  

• [None][fix] Kimi K3 KDA prefill: K1 warps finish reading sPartialLast before the next chunk

  Community want to contribute
  #19826·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  1

  

• [None][perf] Enable tiled FP8 context FMHA for head_dim 256 on SM120

  #19825·amukkara opened Oct 2, 2026Collaborator·1 task done·

  ·
  1

  

• [None][feat] RI-02 bind shared KV leases to lifecycle retirement

  #19824·chienchunhung opened Oct 2, 2026Collaborator·1 task done·

  ·
  3

  

• [None][feat] RI-01 pin shared KV contracts and staging compatibility

  #19823·chienchunhung opened Oct 2, 2026Collaborator·1 task done·

  ·
  3

  

• [None][chore] Share KV transfer lifecycle enforcement primitives

  #19822·chienchunhung opened Oct 2, 2026Collaborator·1 task done·

  ·
  4

  

• [TRTLLM-16282][feat] Add sparse KV offload and GPU metadata publication

  api-compatible
  #19821·cascade812 opened Oct 2, 2026Collaborator·4 of 5 tasks·

  ·
  

• [TRTLLM-14620][feat] Add DSpark confidence verification policy

  #19818·zhaoyuanh-nvidia opened Oct 2, 2026Collaborator·5 of 7 tasks·

  ·
  4

  

• [None][feat] KDA verify: optional per-token states in the V2 hybrid cache manager

  Community want to contribute
  #19817·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  6

  

• [None][perf] Cut host work between speculative decoding steps

  Community want to contribute
  #19816·vsabavat opened Oct 2, 2026·8 of 10 tasks·

  ·
  3