Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on May 19, 2025

• add back DeepRetrieval and add a new work s3

  pat-jj

  committed
  edc3d0dView commit details

  Copy full SHA for edc3d0d

  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

• [misc] fix: no need to use world_size to decide whether to use full_tensor in FSDP2 (#1529)

  Show description for 1466760
  PeterSH6

  authored
  1466760View commit details

  Copy full SHA for 1466760

  Browse repository at this point

• Add Seed-Coder project in README (#1532)

  Show description for 11622fc
  yuyuz

  authored
  11622fcView commit details

  Copy full SHA for 11622fc

  Browse repository at this point

• [feat] Sandbox: support sandbox fusion on FaaS & localhost (#1429)

  Show description for 2c8b2b9
  chenhaiq

  authored
  2c8b2b9View commit details

  Copy full SHA for 2c8b2b9

  Browse repository at this point

• [lr_schedular] fix: implement proper min_lr_ratio support in cosine scheduler (#1400)

  Show description for e12edc7
  0x404

  authored
  e12edc7View commit details

  Copy full SHA for e12edc7

  Browse repository at this point

Commits on May 14, 2025

• [bugfix] correct retrieval of `max_position_embeddings` from config (#1520)

  llkn-2

  authored
  5370035View commit details

  Copy full SHA for 5370035

  Browse repository at this point

• [doc] update docs for custom tool config (#1523)

  Show description for 2d16173
  Necolizer

  authored
  2d16173View commit details

  Copy full SHA for 2d16173

  Browse repository at this point

• Skip max_position_embeddings > sequence length check for vLLM rollouts if RoPE scaling is used (#1522)

  Show description for 9b45fc1
  mansoor-s

  authored
  9b45fc1View commit details

  Copy full SHA for 9b45fc1

  Browse repository at this point

• [metrics] Add diversed reduce metrics method according to key name (#1497)

  Show description for 258a0d9
  SwordFaith

  authored
  258a0d9View commit details

  Copy full SHA for 258a0d9

  Browse repository at this point

• [Doc/Docker Image] Update mcore image to use vLLM which support qwen3 and rewrite installation from conda (#1505)

  Show description for 43782a2
  ETOgaosion

  authored
  43782a2View commit details

  Copy full SHA for 43782a2

  Browse repository at this point

• Add Absolute Zero to awesome work list (#1514)

  Show description for 0a4d545
  Andrewzh112

  authored
  0a4d545View commit details

  Copy full SHA for 0a4d545

  Browse repository at this point

• [fix][DataProto] Make classmethod `from_single_dict` return a cls not the class name (#1509)

  Show description for 21e3acd
  hongpeng-guo

  authored
  21e3acdView commit details

  Copy full SHA for 21e3acd

  Browse repository at this point

Commits on May 13, 2025

• [utils] Enrich and fix utils from `fsdp_utils` and `seqlen_balancing` (#1495)

  Show description for d4a11eb
  hongpeng-guo

  authored
  d4a11ebView commit details

  Copy full SHA for d4a11eb

  Browse repository at this point

• [doc] Clarifying `gpu_memory_utilization` for different engines (#1491)

  Show description for 9a956c0
  dirtyDan0

  authored
  9a956c0View commit details

  Copy full SHA for 9a956c0

  Browse repository at this point

• [refactor][single_controller] Small refactor and fixes in `worker.py` and `ray.base.py` (#1470)

  Show description for 0338531
  hongpeng-guo

  authored
  0338531View commit details

  Copy full SHA for 0338531

  Browse repository at this point

Commits on May 12, 2025

• [Bug] Fix the problem of long inference timeouts when using Async rollout (#1483)

  Show description for cb1adda
  U-rara

  authored
  cb1addaView commit details

  Copy full SHA for cb1adda

  Browse repository at this point