Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on May 21, 2025

• Simplify FSDP SGLang pre and post process (#1609)

  Show description for 8970cb0
  hebiao064

  authored
  8970cb0View commit details

  Copy full SHA for 8970cb0

  Browse repository at this point

• [constant scheduler] fix: model won't be updated on first training step (#1463)

  0x404

  authored
  80af51bView commit details

  Copy full SHA for 80af51b

  Browse repository at this point

• [megatron] support megatron expert parallel (#1467)

  Show description for add17f0
  

  ISEEKYAN

  and

  ETOgaosion

  authored
  add17f0View commit details

  Copy full SHA for add17f0

  Browse repository at this point

• [Docker Image] update images and fix sglang installation (#1606)

  Show description for 7b0426a
  ETOgaosion

  authored
  7b0426aView commit details

  Copy full SHA for 7b0426a

  Browse repository at this point

• docs: update adoption and doc index (#1607)

  Show description for d135072
  eric-haibin-lin

  authored
  d135072View commit details

  Copy full SHA for d135072

  Browse repository at this point

Commits on May 20, 2025

• Remove unnecessary broadcast (#1597)

  Show description for 0d3360a
  hebiao064

  authored
  0d3360aView commit details

  Copy full SHA for 0d3360a

  Browse repository at this point

• [misc] docs: add design doc of single_controller (#1549)

  Show description for f41a57a
  zw0610

  authored
  f41a57aView commit details

  Copy full SHA for f41a57a

  Browse repository at this point

• [BugFix] Megatron: fix missing grad_norm and lr calculation, and fix fsdp grad_norm storage (#1601)

  Show description for 1d4e23a
  ETOgaosion

  authored
  1d4e23aView commit details

  Copy full SHA for 1d4e23a

  Browse repository at this point

• Support multi-turn rollout with Qwen chat template (#1593)

  Show description for 15b1b15
  jybsuper

  authored
  15b1b15View commit details

  Copy full SHA for 15b1b15

  Browse repository at this point

• Disable fused kernels in prime (#1598)

  Show description for a3c4cb3
  plutoZZZZ

  authored
  a3c4cb3View commit details

  Copy full SHA for a3c4cb3

  Browse repository at this point

• [rollout] perf: replace AsyncOpenAI to aiohttp client in ChatCompletionScheduler (#1588)

  Show description for 3eaaf24
  wuxibin89

  authored
  3eaaf24View commit details

  Copy full SHA for 3eaaf24

  Browse repository at this point

• [feat] support logging to ClearML (#1582)

  Show description for 457ccd9
  pavelgein

  authored
  457ccd9View commit details

  Copy full SHA for 457ccd9

  Browse repository at this point

• [BugFix] Megatron: fix checkpoint manager, some states only rank 0 need to save to avoid errors (#1586)

  Show description for 88527e6
  ETOgaosion

  authored
  88527e6View commit details

  Copy full SHA for 88527e6

  Browse repository at this point

• Bump to sglang 0.4.6.post4 & unified generate sequences ability between sgl and sgl async (#1577)

  Show description for 8160ec6
  

  SwordFaith

  and

  ocss884

  authored
  8160ec6View commit details

  Copy full SHA for 8160ec6

  Browse repository at this point

Commits on May 19, 2025

• [doc] single_controller: Adding doc strings and doc pages for public methods in `single_controller` (#1396)

  Show description for 8788e55
  hongpeng-guo

  authored
  8788e55View commit details

  Copy full SHA for 8788e55

  Browse repository at this point

• README: add back DeepRetrieval and add a new work s3 (#1592)

  Show description for 877e097
  pat-jj

  authored
  877e097View commit details

  Copy full SHA for 877e097

  Browse repository at this point

• [BugFix] fix sglang CI, use stable way to download Qwen 7B model (#1585)

  Show description for ab24d7b
  ETOgaosion

  authored
  ab24d7bView commit details

  Copy full SHA for ab24d7b

  Browse repository at this point

• [misc] use lazy import on megatron utils components (#1551)

  Show description for 5b24e01
  hongpeng-guo

  authored
  5b24e01View commit details

  Copy full SHA for 5b24e01

  Browse repository at this point

• [Megatron] Qwen3moe-part3: fix mcore qwen3 moe config, no need for patching now, offer option to freeze moe router (#1540)

  Show description for 8176d3b
  ETOgaosion

  authored
  8176d3bView commit details

  Copy full SHA for 8176d3b

  Browse repository at this point

• [misc] ci: fix typo in PULL_REQUEST_TEMPLATE.md (#1571)

  zw0610

  authored
  8845a33View commit details

  Copy full SHA for 8845a33

  Browse repository at this point

Commits on May 18, 2025

• [misc] feat: support return full prompt with chat template in RLHFDataset (#1567)

  PeterSH6

  authored
  8653b1bView commit details

  Copy full SHA for 8653b1b

  Browse repository at this point

• disable fused kernels by default (#1568)

  mingruimingrui

  authored
  b9a6890View commit details

  Copy full SHA for b9a6890

  Browse repository at this point

• [merger] fix: avoid setting torch's global device to meta (#1564)

  Show description for 530154e
  0x404

  authored
  530154eView commit details

  Copy full SHA for 530154e

  Browse repository at this point

• Add missing fi to install script (#1559)

  ezyang

  authored
  d36b5e8View commit details

  Copy full SHA for d36b5e8

  Browse repository at this point

Commits on May 17, 2025

• [BUG] Fix silent bug of using dtype from previous loop scope in build_memory_reference_from_module() (#1553)

  Show description for 40dcabe
  ccchow

  authored
  40dcabeView commit details

  Copy full SHA for 40dcabe

  Browse repository at this point

Commits on May 16, 2025

• [Docker Image] use latest vLLM (0.8.5) to fully support Qwen3 moe (#1544)

  ETOgaosion

  authored
  b8bd596View commit details

  Copy full SHA for b8bd596

  Browse repository at this point

• [model merger] refactor model merger for better usage and maintainability (#1468)

  Show description for 3f4647f
  0x404

  authored
  3f4647fView commit details

  Copy full SHA for 3f4647f

  Browse repository at this point

• Feat/memory optimized loss (#1212)

  Show description for eb077f6
  

  mingruimingrui

  and

  ETOgaosion

  authored
  eb077f6View commit details

  Copy full SHA for eb077f6

  Browse repository at this point

• [megatron] Qwen3moe-part 2: Allow Infer and train tp to be different with CI tests, Fix vllm resharding process (#1444)

  Show description for b529564
  ETOgaosion

  authored
  b529564View commit details

  Copy full SHA for b529564

  Browse repository at this point

• [Refactor] Add middle truncation (#1488)

  Show description for 12bb857
  

  GaokaiZhang

  and

  Wangmerlyn

  authored
  12bb857View commit details

  Copy full SHA for 12bb857

  Browse repository at this point

• [megatron] fix head_dim in GQA model when load from hf ckpt (#1513)

  Show description for 2c991f6
  ShareLer

  authored
  2c991f6View commit details

  Copy full SHA for 2c991f6

  Browse repository at this point

• [misc] docs: move dev folder to scripts. add sandbox documentation to index.rst. (#1539)

  Show description for 771bd75
  eric-haibin-lin

  authored
  771bd75View commit details

  Copy full SHA for 771bd75

  Browse repository at this point

Commits on May 15, 2025

• [chore] refactor: clean utils code. (#1290)

  Show description for a43db53
  

  湛露先生

  authored
  a43db53View commit details

  Copy full SHA for a43db53

  Browse repository at this point

• Fix reinforce_plus_plus_baseline advantage mask (#1527)

  hijkzzz

  authored
  4e9586aView commit details

  Copy full SHA for 4e9586a

  Browse repository at this point

• fix #1534, sglang_async missing offload_param config (#1536)

  Show description for 6de40fc
  linxxx3

  authored
  6de40fcView commit details

  Copy full SHA for 6de40fc

  Browse repository at this point