All pull requests

Pull requests

• Add full_recompute_input to fine-grained activation offloading

  community-request
  #8300·avigyabb opened Oct 16, 2026·4 of 7 tasks·

  ·
  1

• Shard KPool indexer selection across tensor-parallel ranks

  community-request
  #8299·avigyabb opened Oct 16, 2026·4 of 7 tasks·

  ·
  1

• Add bf16 reduce-scatter option for 2-rank expert-TP MoE combine

  community-request
  #8298·avigyabb opened Oct 16, 2026·4 of 7 tasks·

  ·
  1

• Release router probs in the all-to-all dispatcher after combine

  community-request
  #8297·avigyabb opened Oct 16, 2026·5 of 7 tasks·

  ·
  1

• Optimize sigmoid MoE inference routing with bit-exact outputs

  community-request
  #8296·Waterpine opened Oct 16, 2026·3 of 7 tasks·

  ·
  2

• Support standard DCP save/load with MFSDP v2 state-dict hooks

  #8295·wujingyue opened Oct 16, 2026Contributor·3 of 5 tasks·

  ·
  1

• Chenshuz/gigatoken integration

  complexity: medium
  #8294·jupiterepoch opened Oct 16, 2026Contributor·6 tasks done·

  ·
  3

• Fix duplicate parameter all-gathers across MoE distributed optimizers

  community-request
  #8292·ewan0x79 opened Oct 16, 2026·3 of 7 tasks·

  ·
  1

• fix(inference): support request-local FlashInfer sampling seeds

  #8290·wdykas opened Oct 16, 2026Contributor·1 task·

  ·
  1

• ci(notifications): alert on merge-group testing failures

  complexity: low
  #8289·ko3n1g opened Oct 16, 2026Contributor·

  ·
  2

• fix(tests): assert restored tensor values in test_empty_load

  community-request
  #8288·LZC-BELIEVER opened Oct 16, 2026·3 of 4 tasks·

  ·
  2

• [Dev] Add KPool DSA indexer, NoPE MLA support, and FP8 precision controls for the indexer and MLA attention (dev port of #7522)

  #8287·buptzyb opened Oct 16, 2026Contributor·3 tasks done·

  ·
  2

• Honor per-module quantization override in the fused GroupedMLP path

  #8286·xrennvidia opened Oct 16, 2026Contributor·1 of 7 tasks·

  ·
  2

• [dev] Mask padding before HybridEP dispatch under an expert-rank budget

  #8285·xiaoyao0115 opened Oct 16, 2026Contributor·3 of 6 tasks·

  ·
  1

• [MFSDP v2] Record iteration demand order for VPP and combined 1F1B

  #8284·shjwudp opened Oct 16, 2026Contributor·1 of 3 tasks·

  ·
  1

• Add opt-in expert deployment layout exchange to mLite

  community-request
  #8283·lidongdongbuaa opened Oct 16, 2026·5 of 7 tasks·

  ·
  1

• migrate GDP and its CP implementation to cudnn frontend

  #8282·Mellonta opened Oct 16, 2026Contributor·7 tasks·

  ·
  1

• feat(moe): Qwen3.5 support and speedups for the MoK megakernel backend

  #8281·tomlifu opened Oct 15, 2026Contributor·5 of 13 tasks·

  ·
  1

  

• [otel-7/7] Report structured GPU-sniff measurements through the telemetry adapter

  #8280·rhewett-nv opened Oct 15, 2026Contributor·7 tasks done·

  ·
  1

• [otel-6/7] Count global tokens at the committed training-work boundary

  #8279·rhewett-nv opened Oct 15, 2026Contributor·7 tasks done·

  ·
  1

• [otel-5/7] Emit training report events from existing host scalar values

  #8278·rhewett-nv opened Oct 15, 2026Contributor·7 tasks done·

  ·
  1

• [otel-4/7] Own startup, linked training traces, and exit cleanup in telemetry

  #8277·rhewett-nv opened Oct 15, 2026Contributor·7 tasks done·

  ·
  1

• [otel-3/7] Use canonical MCore spans at production call sites

  #8276·rhewett-nv opened Oct 15, 2026Contributor·7 tasks done·

  ·
  1

• [otel-2/7] Define canonical training spans and stable iteration identity

  #8275·rhewett-nv opened Oct 15, 2026Contributor·7 tasks done·

  ·
  1

• Fix padding-token zeroing for quantized inference under CUDA graphs

  Approvedcomplexity: mediumdeterminism-exemptRun functional tests
  #8274·ilml opened Oct 15, 2026Contributor·4 of 7 tasks·

  ·
  2