All pull requests

Pull requests

• chore(beep boop 🤖): Bump uv.lock (core_r0.20.0) (2026-10-19)

  Run functional tests
  #8337·svcnvidia-nemo-ci opened Oct 19, 2026Contributor·1 task·

  ·
  2

• chore(beep boop 🤖): Bump uv.lock (main) (2026-10-19)

  Run functional tests
  #8336·svcnvidia-nemo-ci opened Oct 19, 2026Contributor·1 task·

  ·
  3

• Fix YaRN cache reuse for shorter context-parallel sequences

  community-request
  #8335·Sunt-ing opened Oct 19, 2026Contributor·4 of 7 tasks·

  ·
  1

• refactor(checkpoint): re-architect CheckpointWithoutOutput internals

  #8334·hxbai opened Oct 19, 2026Contributor·7 tasks·

  ·
  1

• [fsdp, tensor_parallel] fix: preserve local linear gradient accumulation

  community-request
  #8333·tang-t21 opened Oct 19, 2026·6 of 7 tasks·

  ·
  1

  2

• Avoid unused output weights on tied MTP stages

  complexity: low
  #8330·shjwudp opened Oct 19, 2026Contributor·

  ·
  1

  

• [MFSDP v2] Reuse local gradients across reduce-scatter windows

  #8329·shjwudp opened Oct 19, 2026Contributor·

  ·
  1

• Fix pad_routing_map capacity checks for unaligned expert counts

  community-request
  #8328·Eliaaazzz opened Oct 19, 2026·3 of 7 tasks·

  ·
  1

  1

• Fix MLA request-arrival consistency in batch-invariant inference

  community-request
  #8327·Sunt-ing opened Oct 19, 2026Contributor·7 tasks done·

  ·
  1

• Select the NCCL symmetric-memory backend before every MFSDP v2 test

  complexity: lowRun tests
  #8326·ilml opened Oct 19, 2026Contributor·3 of 7 tasks·

  ·
  2

• cp: use GitHub App authentication for code freeze

  Approved
  #8323·svcnemo-autobot opened Oct 18, 2026Collaborator·

  ·
  3

• Fix Torch GDN recurrent state allocation during CUDA graph capture

  26.10cherry-pickcomplexity: lowcore_r0.20.0Final Review
  #8322·cuichenx opened Oct 18, 2026Contributor·2 of 4 tasks·

  ·
  2

• Fix default dense TP reduction in batch-invariant mode

  community-request
  #8321·Sunt-ing opened Oct 18, 2026Contributor·5 of 7 tasks·

  ·
  1

• Optimize MoE expert compute

  community-request
  #8320·Waterpine opened Oct 18, 2026·4 of 7 tasks·

  ·
  3

• fix(inference): promote TorchSampling temperature divisor to FP32

  community-request
  #8316·muditagrawal2007 opened Oct 17, 2026·

  ·
  1

  1

• fix(optimizer): skip redundant state reload in DistributedOptimizer.load_state_dict

  community-request
  #8315·muditagrawal2007 opened Oct 17, 2026·

  ·
  1

  1

• Add output-layer activation gradient clipping

  complexity: low
  #8314·mkhona-nvidia opened Oct 17, 2026Contributor·1 task·

  ·
  1

• MFSDP v2: add Muon with sharded momentum

  complexity: medium
  #8313·wujingyue opened Oct 17, 2026Contributor·1 task done·

  ·
  15

• [MFSDP] Support shortcut MoE blocks and split residual units

  complexity: medium
  #8312·wujingyue opened Oct 17, 2026Contributor·3 of 6 tasks·

  ·
  1

• revert(multimodal): AUT-2553 remove obsolete VLM evaluation benchmarks

  Run full unit tests
  #8311·svcnemo-autobot opened Oct 17, 2026Collaborator·

  ·
  4

• fix(core): compare major version in experimental EOL check

  community-request
  #8309·NUAAwqy opened Oct 17, 2026·

  ·
  2

• Support GDN2 and TE CUDA graphs for recurrence offloading

  community-request
  #8308·Pinoeer-kingxi opened Oct 17, 2026·6 of 7 tasks·

  ·
  1

• Optimize NCCL MoE dispatch

  community-request
  #8307·Waterpine opened Oct 17, 2026·4 of 7 tasks·

  ·
  3

• Optimize NVLS MoE dispatch

  community-request
  #8306·Waterpine opened Oct 17, 2026·4 of 7 tasks·

  ·
  3

• chore: AUT-2547 nightly sync main into dev (2026-10-16)

  Run functional tests
  #8305·svcnemo-autobot opened Oct 17, 2026Collaborator·

  ·
  13