Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Jul 8, 2026

• [BREAKING][trainer] fix: separate_async should use the same step granularity with other trainers (#6977)

  Show description for a5ca129
  Begunner

  authored
  a5ca129View commit details

  Copy full SHA for a5ca129

  View code at this pointBrowse repository at this point

• [rollout] feat: vLLM Prefill-Decode disaggregated rollout (NIXL + Mooncake wiring) (#6243)

  Show description for 828bd39
  

  5 peopleauthored
  828bd39View commit details

  Copy full SHA for 828bd39

  View code at this pointBrowse repository at this point

• [trainer] fix: only set CUDA_DEVICE_MAX_CONNECTIONS=1 for megatron in Hopper/Ampere (#6976)

  Show description for e0275fc
  wuxibin89

  authored
  e0275fcView commit details

  Copy full SHA for e0275fc

  View code at this pointBrowse repository at this point

• [env] chore: Adapt qwen3.5 to Docker containers for Verl release 0.8.0 (#6955)

  Show description for 5fc49b7
  ruanhao566

  authored
  5fc49b7View commit details

  Copy full SHA for 5fc49b7

  View code at this pointBrowse repository at this point

Commits on Jul 7, 2026

• [ckpt] fix: use separate magic_recv buffer to prevent weight corruption (#6813)

  Show description for 7b19203
  KunWuLuan

  authored
  7b19203View commit details

  Copy full SHA for 7b19203

  View code at this pointBrowse repository at this point

• [perf] feat: Add Qwen3.5 MFU flops estimation (#6389)

  Show description for dbd307e
  liangxuZhang

  authored
  dbd307eView commit details

  Copy full SHA for dbd307e

  View code at this pointBrowse repository at this point

• [algo] fix: normalize critic value loss over the global mini-batch, not per micro-batch (#6957)

  Show description for 2eb020a
  abtonmoy

  authored
  2eb020aView commit details

  Copy full SHA for 2eb020a

  View code at this pointBrowse repository at this point

• [megatron] fix: free Transformer-Engine FP8 weight workspaces on CPU offload (#6952)

  Show description for e52747a
  

  4 peopleauthored
  e52747aView commit details

  Copy full SHA for e52747a

  View code at this pointBrowse repository at this point

• [fsdp] fix: fused-kernel gradient tracking bug. (#6913)

  Show description for e48725a
  

  kolehma8

  and

  gemini-code-assist[bot]

  authored
  e48725aView commit details

  Copy full SHA for e48725a

  View code at this pointBrowse repository at this point

• [vllm, rollout] fix: require vLLM>=0.22.0 for Qwen3.5 router replay (R3) (#6922)

  Show description for 2ef858e
  dafu-wu

  authored
  2ef858eView commit details

  Copy full SHA for 2ef858e

  View code at this pointBrowse repository at this point

• [trainer] fix: spmd_types and activation checkpointing composability bug (#6950)

  Show description for ef500de
  acisseJZhong

  authored
  ef500deView commit details

  Copy full SHA for ef500de

  View code at this pointBrowse repository at this point

Commits on Jul 6, 2026

• [hardware, perf] fix: update AMD GPU FLOPS values (#6942)

  Show description for 9481350
  Vivicai1005

  authored
  9481350View commit details

  Copy full SHA for 9481350

  View code at this pointBrowse repository at this point

• [rollout] fix: probe list-of-parts content in initialize_turn_separator for multimodal processors (#6937)

  Show description for e3d9bad
  abtonmoy

  authored
  e3d9badView commit details

  Copy full SHA for e3d9bad

  View code at this pointBrowse repository at this point

• [fsdp] fix: FSDP logits temperature scaling for view tensors (#6935)

  Show description for 51504d8
  zjchenn

  authored
  51504d8View commit details

  Copy full SHA for 51504d8

  View code at this pointBrowse repository at this point

• [vllm, rollout] fix: handle non-contiguous weights in bucketed transfer (#6919)

  Show description for c47d54e
  Mecoli1219

  authored
  c47d54eView commit details

  Copy full SHA for c47d54e

  View code at this pointBrowse repository at this point

• [rollout] fix: restore turn separator dropped at multi-turn tool agent loop boundaries (#6921)

  Show description for 6874ecd
  abtonmoy

  authored
  6874ecdView commit details

  Copy full SHA for 6874ecd

  View code at this pointBrowse repository at this point

• [trainer] fix: Update latest TorchtitanEngine (#6916)

  Show description for d78305d
  acisseJZhong

  authored
  d78305dView commit details

  Copy full SHA for d78305d

  View code at this pointBrowse repository at this point

• [megatron] fix: guard ModelType.encoder_and_decoder for Megatron-Core >= 0.18 compatibility (#6927)

  Show description for bb4cd0a
  chengcuiping

  authored
  bb4cd0aView commit details

  Copy full SHA for bb4cd0a

  View code at this pointBrowse repository at this point

• [fully_async] feat: Adapt vLLM >=0.18 and vLLM < 0.19 for Ascend NPU (#6929)

  Show description for 45a37f8
  

  zhouhengan1211

  and

  zhouhengan

  authored
  45a37f8View commit details

  Copy full SHA for 45a37f8

  View code at this pointBrowse repository at this point

Commits on Jul 2, 2026

• [fsdp] fix: Fix Qwen3 MoE FSDP weight sync for vLLM rollout in Transformers 5 (#6896)

  Show description for 8c2fa63
  lxb007981

  authored
  8c2fa63View commit details

  Copy full SHA for 8c2fa63

  View code at this pointBrowse repository at this point

• [rollout] fix: support SGLang FP8 ignored layers for Qwen3.x GatedDeltaNet in rollout (#6906)

  Show description for 91666d9
  

  gem-mint

  and

  codex

  authored
  91666d9View commit details

  Copy full SHA for 91666d9

  View code at this pointBrowse repository at this point

• [megatron, perf] feat: pad BSHD micro-batches to mini-batch max seq_len (#6901)

  Show description for 561a4b0
  YaoweiFan

  authored
  561a4b0View commit details

  Copy full SHA for 561a4b0

  View code at this pointBrowse repository at this point

Commits on Jul 1, 2026

• [fully_async] feat: Adapt vLLM 0.19+ for Ascend NPU (#6886)

  Show description for 7dc5559
  fh188

  authored
  7dc5559View commit details

  Copy full SHA for 7dc5559

  View code at this pointBrowse repository at this point

• [data, rollout, worker] feat: add Open-R1 multimodal and TinyLLaVA-Video-R1 preprocessing and training scripts (#6849)

  Show description for 6751367
  lihanwen7

  authored
  6751367View commit details

  Copy full SHA for 6751367

  View code at this pointBrowse repository at this point

Commits on Jun 30, 2026

• [fsdp, megatron, trainer] fix: enhance mem footprint for forward_kl_topk OPD (#6848)

  Show description for 1e3b6ff
  

  dimjava

  and

  Dmitrii Choklia

  authored
  1e3b6ffView commit details

  Copy full SHA for 1e3b6ff

  View code at this pointBrowse repository at this point

• [vllm] fix: crash in start_profile/stop_profile on non-master nodes when nnodes > 1 (#6861)

  Show description for 2c7e3f3
  kyle-zhangchi

  authored
  2c7e3f3View commit details

  Copy full SHA for 2c7e3f3

  View code at this pointBrowse repository at this point

• [hardware] fix: add flops estimation for ascend a2g3 device (#6892)

  Show description for a7f5321
  Mengyuyang

  authored
  a7f5321View commit details

  Copy full SHA for a7f5321

  View code at this pointBrowse repository at this point

Commits on Jun 29, 2026

• [fully_async, doc] fix: ignore temperature config for teacher prompt_logprobs and warn when non-default value is set (#6867)

  Show description for c165b3d
  

  kenkenpa2126

  and

  claude

  authored
  c165b3dView commit details

  Copy full SHA for c165b3d

  View code at this pointBrowse repository at this point

• [model] fix: tokenizer.hf_processor: GLM-4V arm uses wrong processor class name (#6814) (#6873)

  Show description for 7ba7a5d
  yushuosun

  authored
  7ba7a5dView commit details

  Copy full SHA for 7ba7a5d

  View code at this pointBrowse repository at this point

• [fully_async] fix: correct the use of partial_rollout (#6882)

  Show description for 5371b99
  zpltys

  authored
  5371b99View commit details

  Copy full SHA for 5371b99

  View code at this pointBrowse repository at this point

• [trainer, rollout] feat: log rollout moe load-balance metrics (#6853)

  Show description for 3d66a3d
  Luosuu

  authored
  3d66a3dView commit details

  Copy full SHA for 3d66a3d

  View code at this pointBrowse repository at this point

Commits on Jun 27, 2026

• [worker] fix: Update external_lib config annotation in HFModelConfig to allow a list of string (#6862)

  Show description for 1ff76cc
  

  wyettzeng

  and

  huaye.zeng

  authored
  1ff76ccView commit details

  Copy full SHA for 1ff76cc

  View code at this pointBrowse repository at this point

Commits on Jun 26, 2026

• [fsdp] fix: re-wrap fused log_probs/entropy when `use_remove_padding=False` (#6846)

  Show description for 2390a3f
  Mecoli1219

  authored
  2390a3fView commit details

  Copy full SHA for 2390a3f

  View code at this pointBrowse repository at this point

• fix typo: clip_cvo_ratio to clip_cov_ratio (#6855)

  Show description for 6fef6a7
  GZYZG

  authored
  6fef6a7View commit details

  Copy full SHA for 6fef6a7

  View code at this pointBrowse repository at this point

• [rollout, algo] fix: compute rollout_is_seq_fraction from raw weights (#6806)

  Show description for bdfedd4
  

  EazyReal

  and

  claude

  authored
  bdfedd4View commit details

  Copy full SHA for bdfedd4

  View code at this pointBrowse repository at this point