Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit History

Commits on Jun 12, 2026

• [reward] fix: add optional per-sample compute_score timeout to NaiveRewardManager (#6673)

  Show description for 9f18abe
  zqinglin

  authored
  9f18abe

  Copy full SHA for 9f18abe

• [hardware] feat: add ROCm/HIP platform backend (PlatformROCm) (#6702)

  Show description for ab8da13
  

  xiaohong42

  and

  cursoragent

  authored
  ab8da13

  Copy full SHA for ab8da13

• [rollout] fix: clone LoRA weights out of the reused IPC buffer before add_lora (#6688)

  Show description for 173945f
  

  HaozheZhang6

  and

  claude

  authored
  173945f

  Copy full SHA for 173945f

• [fsdp] fix: handle missing chunked top-k config (#6704)

  Show description for 4a0ff06
  Luosuu

  authored
  4a0ff06

  Copy full SHA for 4a0ff06

• [fsdp, trainer] fix: detach model_output and loss metrics to stop per-micro-batch graph retention in actor update (#6699)

  Show description for dbbf085
  YoungZSh

  authored
  dbbf085

  Copy full SHA for dbbf085

Commits on Jun 10, 2026

• [megatron] fix: propagate disabled MTP config to bridge providers (#6626)

  Show description for 41a5244
  Mengyuyang

  authored
  41a5244

  Copy full SHA for 41a5244

• [rollout] fix: propagate ignore_eos to vLLM sampling params (#6630)

  Show description for 15560ea
  Mengyuyang

  authored
  15560ea

  Copy full SHA for 15560ea

• [trainer, megatron] fix: fix MindSpeed repatch config bug (#6674)

  Show description for a1fad97
  chengminhua

  authored
  a1fad97

  Copy full SHA for a1fad97

• [fsdp] feat: chunked gather-logsumexp for top-K loss to avoid OOM at long context (#6593)

  kekellllll

  authored
  c6144ec

  Copy full SHA for c6144ec

• [veomni, cfg] feat: add missing config fields to veomni.yaml (#6505)

  Show description for 5f74f82
  mikequan0425

  authored
  5f74f82

  Copy full SHA for 5f74f82

Commits on Jun 9, 2026

• [vllm, fsdp] fix: apply FSDP buffer updates during rollout weight sync (#5801)

  Show description for a1f6e71
  chenshui223

  authored
  a1f6e71

  Copy full SHA for a1f6e71

Commits on Jun 8, 2026

• [vllm] fix: use data-parallel rank in vLLM ZMQ handles (#6620)

  Show description for 478b718
  he-yufeng

  authored
  478b718

  Copy full SHA for 478b718

• [megatron] fix: MTP compatible with latest mcore (#6648)

  Show description for f65b2ce
  HollowMan6

  authored
  f65b2ce

  Copy full SHA for f65b2ce

• [data] chore: Update TransferQueue version to 0.1.8 (#6652)

  Show description for cea8661
  0oshowero0

  authored
  cea8661

  Copy full SHA for cea8661

• [data] fix: support path multimodal placeholders (#6631)

  Show description for f48b143
  le-czs

  authored
  f48b143

  Copy full SHA for f48b143

• [megatron] fix: Fix Megatron optimizer offloading when `use_precision_aware_optimizer=True` (#6635)

  Show description for 587b120
  

  rao-ashish

  and

  gemini-code-assist[bot]

  authored
  587b120

  Copy full SHA for 587b120

• [megatron] fix: official moe_enable_routing_replay flag in addition (#6647)

  Show description for 62197aa
  HollowMan6

  authored
  62197aa

  Copy full SHA for 62197aa

• [veomni] fix: pass mixed precision config to veomni (#6644)

  Show description for 9add650
  

  Luosuu

  and

  codex

  authored
  9add650

  Copy full SHA for 9add650

Commits on Jun 7, 2026

• [rollout] fix: avoid arbitrary code execution in Qwen3 tool parser (#6542)

  Show description for 2084d8f
  

  fuergaosi233

  and

  claude

  authored
  2084d8f

  Copy full SHA for 2084d8f

Commits on Jun 5, 2026

• [megatron] fix: clamp num_tokens=0 in MTP loss & add normalized scale for MTP per token loss (#6464)

  Show description for 8911f24
  arvyanh

  authored
  8911f24

  Copy full SHA for 8911f24

• [fsdp] fix: do not manually move model to GPU (#6604)

  Show description for a539474
  

  pavelgein

  and

  gemini-code-assist[bot]

  authored
  a539474

  Copy full SHA for a539474

• [rollout] chore: change default to calculate_log_probs=True (#6616)

  Show description for 614ffc0
  wuxibin89

  authored
  614ffc0

  Copy full SHA for 614ffc0

• [data] fix: default image_patch_size to processor's real patch_size in filter_overlong_prompts (#6595)

  Show description for f1478d2
  

  zengxingchen

  and

  gemini-code-assist[bot]

  authored
  f1478d2

  Copy full SHA for f1478d2

• [veomni] feat: reduce extra memory from O(E) to O(E/ep_size) during weight refit (#6612)

  Show description for c2d93fc
  wuxibin89

  authored
  c2d93fc

  Copy full SHA for c2d93fc

• [perf] feat: support partial-token window profiling for rollout in verl (#5760)

  Show description for 381ed24
  mengchengTang

  authored
  381ed24

  Copy full SHA for 381ed24

Commits on Jun 4, 2026

• [hardware] feat: add platform abstraction layer and plugin-based engine override system (#6086)

  Show description for 7a457c9
  

  3 peopleauthored
  7a457c9

  Copy full SHA for 7a457c9

• [vllm, megatron] fix: mxfp8 trainning support on Ascend NPU (#6562)

  Show description for 0ce9478
  

  quancs

  and

  gemini-code-assist[bot]

  authored
  0ce9478

  Copy full SHA for 0ce9478

• [tool, rollout] feat: Use SkipManager to uniformly manage the skipping schemes. (#6097)

  Show description for 266fdce
  

  3 peopleauthored
  266fdce

  Copy full SHA for 266fdce

Commits on Jun 2, 2026

• [megatron, ckpt] fix: handle None param_data in get_megatron_module_device when use_distributed_optimizer=False (#5782)

  Show description for c34c001
  khazic

  authored
  c34c001

  Copy full SHA for c34c001

• [megatron, ckpt] refactor: optimize mcore ckpt manager impl, checkpoint yaml config APIs (#6014)

  Show description for 22b04f9
  ETOgaosion

  authored
  22b04f9

  Copy full SHA for 22b04f9

• [vllm] fix: reset all caches after weight updates (#6522)

  Show description for accb1a0
  s-isaev

  authored
  accb1a0

  Copy full SHA for accb1a0

• [reward] fix: release worker count when NaiveRouter request fails after retries (#6544)

  Show description for 11ecc15
  

  fuergaosi233

  and

  claude

  authored
  11ecc15

  Copy full SHA for 11ecc15

Commits on Jun 1, 2026

• [megatron] fix: Fix GPU memory leak in ref model offload by explicitly releasing storage (#6447)

  Show description for 0a1c5b6
  ZLiao097

  authored
  0a1c5b6

  Copy full SHA for 0a1c5b6

• [misc] chore: bump version to 0.9.0.dev (#6553)

  Show description for 8f5e161
  wuxibin89

  authored
  8f5e161

  Copy full SHA for 8f5e161

• [ci] fix: FSDP actor/critic ci fail (#6550)

  Show description for 7aed6b2
  wuxibin89

  authored
  7aed6b2

  Copy full SHA for 7aed6b2