All pull requests

Pull requests

• refactor(dtensor): delete the DTensor v1 native checkpoint implementation

  CI:Lfast
  #4273·yuki-97 opened Sep 25, 2026Contributor·

  ·
  4

• refactor(config): centralize the outdated-config checks in one module

  CI:L1
  #4271·yuki-97 opened Sep 25, 2026Contributor·2/2·

  ·
  14

• chore(dtensor): delete the dtensor_cfg._v2 key

  CI:Lfast
  #4270·yuki-97 opened Sep 25, 2026Contributor·1/2·

  ·
  5

• ci: Bump Megatron-Bridge to 4566ea8

  CI:L1
  #4269·svcnvidia-nemo-ci opened Sep 25, 2026Contributor·

  ·
  3

• feat: track Bridge-preferred MFU in SFT v2 and single-controller

  CI:L1Documentation
  #4268·terrykong opened Sep 25, 2026Collaborator·

  ·
  15

• feat(checkpoint): orchestrate Gym turn-level recovery

  Documentation
  #4266·macandro96 opened Sep 25, 2026Contributor·3/3·

  ·
  1

• feat(checkpoint): persist recoverable Gym rollout state

  #4265·macandro96 opened Sep 25, 2026Contributor·2/3·

  ·
  1

• feat(checkpoint): add Gym participant coordination

  #4264·macandro96 opened Sep 25, 2026Contributor·1/3·

  ·
  1

• fix(data-plane): preserve tensor layouts and avoid int16 widening

  community-request
  #4263·Alvorecer721 opened Sep 25, 2026·

  ·
  1

  1

• build(docker): install uv and its managed Python outside /root

  CI:L0
  #4260·ko3n1g opened Sep 24, 2026Contributor·

  ·
  3

• fix(dtensor): merge LoRA into refit weights under checkpoint wrappers

  community-request
  #4259·zhiyuanma-agents opened Sep 24, 2026·

  ·
  1

• feat(research): add Block JustGRPO and Fast Block JustGRPO

  #4258·sajadn opened Sep 24, 2026Contributor·

  ·
  1

• fix(vllm): finalize padded TRTLLM reloads at logical size

  CI:L0
  #4257·seonjinn opened Sep 24, 2026Contributor·

  ·
  1

• feat(ppo): overlap rollout generation with critic training

  Documentation
  #4256·alanliu-nvidia opened Sep 24, 2026Contributor·

  ·
  1

• feat(grpo): support multiple policy updates per rollout

  Documentation
  #4255·sajadn opened Sep 23, 2026Contributor·

  ·
  2

• Nadavsch/argmaxrl

  community-requestDocumentationwaiting-on-maintainers
  #4251·nadavsch-aai opened Sep 23, 2026·

  ·
  2

• ci: bump preflight workflows to v1.11.0

  CI:LfastCI
  #4250·ko3n1g opened Sep 23, 2026Contributor·

  ·
  2

• feat(token-capture): read Gym manifests directly from local ledgers

  CI:L0Documentation
  #4247·pthombre opened Sep 23, 2026Contributor·

  ·
  5

• fix(sc): checkpoint monotonic NeMo Gym task IDs

  CI:L0
  #4245·mehraakash opened Sep 22, 2026Contributor·

  ·
  9

• feat(ppo): stream colocated policy updates in SingleController

  CI:L2Documentation
  #4244·youngeunkwon0405 opened Sep 22, 2026Contributor·

  ·
  12

  

• fix: routed-expert range check and default route in int64 for uint16 ids

  community-request
  #4240·zhiyuanma-agents opened Sep 22, 2026·

  ·
  1

• docs: add NeMo Gym environment guidance

  community-requestDocumentationwaiting-on-maintainers
  #4239·judyheflin opened Sep 22, 2026·

  ·
  1

  1

• fix: Mask generated media toks and pipe style configs to MInf.

  CI:LfastDocumentationinferencetraining
  #4238·cspades opened Sep 21, 2026Contributor·

  ·
  1

  

• feat(dynamo): forward multimodal rollout data

  #4237·pstjohn opened Sep 21, 2026Contributor·

  ·
  1

• feat(recipes): add async SpatialClaw PivotRL

  CI:L1Documentation
  #4236·ehosseiniasl opened Sep 21, 2026Contributor·

  ·
  3