Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Feb 20, 2026

• [trainer] feat: Add Torchtitan as alternative training engine (#5051)

  Show description for f5c34bb
  acisseJZhong

  authored
  f5c34bbView commit details

  Copy full SHA for f5c34bb

  View code at this pointBrowse repository at this point

Commits on Feb 16, 2026

• Revert "[vllm] feat: remove workers from vLLMHttpServer" (#5333)

  Show description for ef26847
  PeterSH6

  authored
  ef26847View commit details

  Copy full SHA for ef26847

  View code at this pointBrowse repository at this point

• [vllm] feat: remove workers from vLLMHttpServer (#5330)

  Show description for 4c9e3f7
  

  tongyx361

  and

  gemini-code-assist[bot]

  authored
  4c9e3f7View commit details

  Copy full SHA for 4c9e3f7

  View code at this pointBrowse repository at this point

Commits on Feb 14, 2026

• [fsdp,algo] feat: add NVFP4 QAT (Quantization-Aware Training) support (#5190)

  Show description for 27f65f1
  zhangyimi

  authored
  27f65f1View commit details

  Copy full SHA for 27f65f1

  View code at this pointBrowse repository at this point

• [veomni, trainer] fix: padding pixel value with padding_scale for vl model (#5322)

  Show description for 8c41724
  

  3 peopleauthored
  8c41724View commit details

  Copy full SHA for 8c41724

  View code at this pointBrowse repository at this point

• [ci] fix: main pre-commit (#5318)

  Show description for 0c5cc48
  pengwu22

  authored
  0c5cc48View commit details

  Copy full SHA for 0c5cc48

  View code at this pointBrowse repository at this point

Commits on Feb 13, 2026

• [megatron] feat: Support LoRA training with FP16 using Megatron-Bridge. (#4648)

  Show description for ec123e6
  xichengpro

  authored
  ec123e6View commit details

  Copy full SHA for ec123e6

  View code at this pointBrowse repository at this point

• [model,cfg] fix: type annotation for Lora target_modules (#5223)

  Show description for 536a978
  

  thvasilo

  and

  gemini-code-assist[bot]

  authored
  536a978View commit details

  Copy full SHA for 536a978

  View code at this pointBrowse repository at this point

• [trtllm] feat: use max utilization scheduler by default (#5302)

  Show description for 4dd4980
  tongyuantongyu

  authored
  4dd4980View commit details

  Copy full SHA for 4dd4980

  View code at this pointBrowse repository at this point

• [veomni] refactor: Modify dp related parameters to align with FSDP backend and remove temporarily unsupported TP/PP/CP parameters (#5303)

  Show description for a43eecb
  

  ChengQianqian

  and

  ji-huazhong

  authored
  a43eecbView commit details

  Copy full SHA for a43eecb

  View code at this pointBrowse repository at this point

Commits on Feb 12, 2026

• [fsdp,veomni] fix: remove FSDPUlyssesShardingManager to make eval_mode/train_mode reentrant (#5305)

  Show description for 9cb89d8
  wuxibin89

  authored
  9cb89d8View commit details

  Copy full SHA for 9cb89d8

  View code at this pointBrowse repository at this point

Commits on Feb 11, 2026

• [BREAKING][reward] refactor: the full reward configuration (#5255)

  Show description for 67dc84b
  yyDing1

  authored
  67dc84bView commit details

  Copy full SHA for 67dc84b

  View code at this pointBrowse repository at this point

• [fsdp, veomni, trainer] fix: restrict npu-patch scope to avoid veomni backend interference (#5268)

  Show description for d643921
  ji-huazhong

  authored
  d643921View commit details

  Copy full SHA for d643921

  View code at this pointBrowse repository at this point

• [vllm, rollout] fix: auto-downgrade cudagraph_mode to PIECEWISE when DCP is enabled (#5262)

  Show description for b6e8357
  Siritao

  authored
  b6e8357View commit details

  Copy full SHA for b6e8357

  View code at this pointBrowse repository at this point

• [rollout] feat: make more rollout flags configurable to trtllm backend (#5258)

  Show description for 2d4ffd4
  Superjomn

  authored
  2d4ffd4View commit details

  Copy full SHA for 2d4ffd4

  View code at this pointBrowse repository at this point

Commits on Feb 10, 2026

• [vllm] fix: run post-load weight processing once after async IPC sync (#5235)

  Show description for 6c5d236
  zjchenn

  authored
  6c5d236View commit details

  Copy full SHA for 6c5d236

  View code at this pointBrowse repository at this point

• [veomni, trainer] feat: add rl support for veomni backend (#4882)

  Show description for 2487c36
  

  ji-huazhong

  and

  A1waysBeenHere

  authored
  2487c36View commit details

  Copy full SHA for 2487c36

  View code at this pointBrowse repository at this point

• [vllm, rollout] feat: update abort function with vllm internal pause_generation api (#5253)

  Show description for 9b70310
  PeterSH6

  authored
  9b70310View commit details

  Copy full SHA for 9b70310

  View code at this pointBrowse repository at this point

Commits on Feb 9, 2026

• [BREAKING][reward] refactor: deprecate batch reward manager (#5237)

  Show description for 3df21a6
  yyDing1

  authored
  3df21a6View commit details

  Copy full SHA for 3df21a6

  View code at this pointBrowse repository at this point

• [algo] feat: reduce routed expert padding via NestedTensor and uint8 dtype (#5240)

  Show description for 3e0d4b1
  xhx1022

  authored
  3e0d4b1View commit details

  Copy full SHA for 3e0d4b1

  View code at this pointBrowse repository at this point

• [vllm] fix: apply moe weight loader patch for standard wight loading (#5234)

  Show description for a5087cb
  zjchenn

  authored
  a5087cbView commit details

  Copy full SHA for a5087cb

  View code at this pointBrowse repository at this point

Commits on Feb 6, 2026

• [worker] feat: Avoid redundant base weight sync when engine doesn't sleep (#5147)

  Show description for 06449b8
  JohnConnor123

  authored
  06449b8View commit details

  Copy full SHA for 06449b8

  View code at this pointBrowse repository at this point

• [fsdp] fix: Support trust_remote_code during FSDP HugginFace checkpoint save (#5200)

  Show description for f245e2d
  

  4 peopleauthored
  f245e2dView commit details

  Copy full SHA for f245e2d

  View code at this pointBrowse repository at this point

• [BREAKING][reward] refactor: remove reward model worker code and invocation (#5194)

  Show description for 4ef45d0
  yyDing1

  authored
  4ef45d0View commit details

  Copy full SHA for 4ef45d0

  View code at this pointBrowse repository at this point

• [recipe] refactor: refactor ray trainer for separate recipe use. (fully async / one step off) (#5184)

  Show description for 32f2d3d
  ArronHZG

  authored
  32f2d3dView commit details

  Copy full SHA for 32f2d3d

  View code at this pointBrowse repository at this point

• [reward] fix: Add assert to prevent reward NaN caused by overlong_cfg.len=0 (#5216)

  Show description for 21f4e49
  ZLiao097

  authored
  21f4e49View commit details

  Copy full SHA for 21f4e49

  View code at this pointBrowse repository at this point

• [perf] feat: add images_seqlens on mfu calculation for engine_worker (#5207)

  Show description for 759eef8
  

  3 peopleauthored
  759eef8View commit details

  Copy full SHA for 759eef8

  View code at this pointBrowse repository at this point

• [trtllm] fix: add synchronization before resume kv_cache to prevent oom in non-leader ranks (#5208)

  Show description for 18523ea
  shuyixiong

  authored
  18523eaView commit details

  Copy full SHA for 18523ea

  View code at this pointBrowse repository at this point

• [algo] feat: support rollout router replay in MegatronEngine (#5185)

  Show description for d987199
  xhx1022

  authored
  d987199View commit details

  Copy full SHA for d987199

  View code at this pointBrowse repository at this point

• [trtllm] fix: reduce peak mem usage during update_weight() (#5212)

  Show description for a8d5bf9
  hchings

  authored
  a8d5bf9View commit details

  Copy full SHA for a8d5bf9

  View code at this pointBrowse repository at this point

Commits on Feb 5, 2026

• [worker, training_utils] fix: Metric Aggregation Across DP Ranks (#5203)

  Show description for 2320603
  JacobHelwig

  authored
  2320603View commit details

  Copy full SHA for 2320603

  View code at this pointBrowse repository at this point

• [vllm] fix: ignore MoE router layers for FP8 quantization (#5107)

  Show description for 597b63f
  zpqiu

  authored
  597b63fView commit details

  Copy full SHA for 597b63f

  View code at this pointBrowse repository at this point

Commits on Feb 4, 2026

• [perf] fix: infer server profiler args fix (#5121)

  Show description for e4915bd
  mengchengTang

  authored
  e4915bdView commit details

  Copy full SHA for e4915bd

  View code at this pointBrowse repository at this point

• [fsdp] feat: Merge lora in fsdp training to speed up rollout (#5115)

  Show description for d6ced80
  

  amzfang

  and

  Copilot

  authored
  d6ced80View commit details

  Copy full SHA for d6ced80

  View code at this pointBrowse repository at this point

• [megatron] feat: fused kernel suppport for new model engine (#5191)

  Show description for e5f5ea6
  HollowMan6

  authored
  e5f5ea6View commit details

  Copy full SHA for e5f5ea6

  View code at this pointBrowse repository at this point