All pull requests

Pull requests

• feat(refactor): refactor for easier usage

  #429·zigzagcai opened May 15, 2025Collaborator·

  ·
  

• Fix npu fa

  #425·SweetStanleyL opened Apr 16, 2025·

  ·
  

• feat(moe): support fused permute when capacity is not none

  #424·blankde opened Apr 9, 2025Collaborator·

  ·
  

• add tp overlap feature

  #416·sallyjunjun opened Feb 18, 2025Collaborator·

  ·
  

• feat(ampipe): impl ampipe

  #406·blankde opened Jan 17, 2025Collaborator·

  ·
  

• Feat/loong train mla

  #400·huangting4201 opened Jan 2, 2025Collaborator·

  ·
  

• feat(checkpoint): support universal checkpoint

  #394·li126com opened Dec 23, 2024Collaborator·

  ·
  

• add use_fp32_logits flag

  #393·KimmiShi opened Dec 20, 2024Collaborator·

  ·
  

• feat(pipeline): add non-p2p-comm support

  #385·mwiacx opened Dec 4, 2024Contributor·

  ·
  

• fix(mlp): enhance mlp_layer_fusion

  #382·yingtongxiong opened Dec 3, 2024Collaborator·

  ·
  

• check CUDA_DEVICE_MAX_CONNECTIONS and TORCH_NCCL_AVOID_RECORD_STREAMS

  #379·sallyjunjun opened Dec 2, 2024Collaborator·

  ·
  2

  

• [Feat] Heterogeneous Code Part 1: Add Model and Module Code for Chameleon Lumina

  #377·zhhsplendid opened Nov 28, 2024·

  ·
  12

  

• fix(monitor): send exception when feishu alert is enable && remove light monitoring address

  #373·JiaoPL opened Nov 20, 2024Collaborator·

  ·
  

• fix llava model device bugs

  #359·hellozmz opened Oct 28, 2024·

  ·
  

• feat(moe): add gshard token rearrange optim

  #352·blankde opened Oct 21, 2024Collaborator·

  ·
  

• feat(moe): support moe zero1 setting

  #350·blankde opened Oct 16, 2024Collaborator·

  ·
  

• feat(moe): add moe async param handler

  #332·blankde opened Sep 13, 2024Collaborator·

  ·
  1

  

• feat(checkpoint): TP recomputation communication optimization

  #275·li126com opened Jul 8, 2024Collaborator·

  ·
  7

  

• Feat/add zeropp

  #256·chrysantd opened Jun 20, 2024·

  ·
  

• feat(simulator): support parallel cost simulator for internevo

  #243·SolenoidWGT opened Jun 5, 2024Contributor·

  ·
  

• add gradient sharding

  #87·ChenQiaoling00 opened Mar 18, 2024Contributor·

  ·