Commits

Branch selector

 ci_async_tp

User selector

All users

Datepicker

All time

Commit History

Commits on Apr 11, 2025

• print device

  kwen2501

  committed
  f812226

  Copy full SHA for f812226

• print device

  kwen2501

  committed
  d7b2220

  Copy full SHA for d7b2220

• debug

  kwen2501

  committed
  63ad3bd

  Copy full SHA for 63ad3bd

• [CI] Re-enable async TP test

  kwen2501

  committed
  d68a110

  Copy full SHA for d68a110

Commits on Apr 10, 2025

• [BE] Chill loggings in checkpoint load (#1089)

  Show description for 34c2346
  kwen2501

  authored
  34c2346

  Copy full SHA for 34c2346

• remove TORCH_NCCL_AVOID_RECORD_STREAMS env var setting (#1088)

  Show description for 681c66f
  tianyu-l

  authored
  681c66f

  Copy full SHA for 681c66f

• [DeepSeek] Add option for `torch._group_mm` (#1086)

  Show description for 5707c3d
  kwen2501

  authored
  5707c3d

  Copy full SHA for 5707c3d

• [llama4] add Grouped GEMM support for MoE (#1084)

  Show description for dbb34cc
  tianyu-l

  authored
  dbb34cc

  Copy full SHA for dbb34cc

• [Flux] Enable FSDP for flux model training (#1074)

  Show description for ab08612
  wwwjn

  authored
  ab08612

  Copy full SHA for ab08612

Commits on Apr 9, 2025

• remove commented code (#1080)

  Show description for 981552e
  danielvegamyhre

  authored
  981552e

  Copy full SHA for 981552e

Commits on Apr 8, 2025

• [PP] Add microbatch_size and remove num_microbatches config (#1073)

  Show description for 5e2033c
  H-Huang

  authored
  5e2033c

  Copy full SHA for 5e2033c

• [PP] Add pipeline_parallel_layers_per_stage config (#1072)

  Show description for 8b737fe
  H-Huang

  authored
  8b737fe

  Copy full SHA for 8b737fe

Commits on Apr 7, 2025

• [Flux] Add script to download autoencoder before training (#1071)

  Show description for 5568431
  wwwjn

  authored
  5568431

  Copy full SHA for 5568431

• add llama4 as an experiment (#1064)

  Show description for e4e4834
  tianyu-l

  authored
  e4e4834

  Copy full SHA for e4e4834

• [DeepSeek] Add CUDA Graph support (#1070)

  Show description for 95d6d4f
  kwen2501

  authored
  95d6d4f

  Copy full SHA for 95d6d4f

• [float8] Make module filter function for float8 training configurable (#1063)

  Show description for a6524d8
  danielvegamyhre

  authored
  a6524d8

  Copy full SHA for a6524d8

• [BE] Move model creation out of generate function (#1069)

  Show description for a5923a0
  kwen2501

  authored
  a5923a0

  Copy full SHA for a5923a0

• Allow logging bespoke metrics (#1061)

  kashif

  authored
  e48704f

  Copy full SHA for e48704f

• [DeepSeek] Move permutation index generation to GPU (#1062)

  Show description for d8d9620
  kwen2501

  authored
  d8d9620

  Copy full SHA for d8d9620

Commits on Apr 5, 2025

• update CI to use pytorch nightly (#1057)

  Show description for a59ed30
  tianyu-l

  authored
  a59ed30

  Copy full SHA for a59ed30

Commits on Apr 4, 2025

• Onboard FLUX model step 1: Adding dataloader, model code and forward pass (#1024)

  Show description for d2b57cf
  wwwjn

  authored
  d2b57cf

  Copy full SHA for d2b57cf

• add async TP test to CI (#1029)

  tianyu-l

  authored
  82ef16a

  Copy full SHA for 82ef16a

Commits on Apr 3, 2025

• [DeepSeek] Add README (#1053)

  Show description for 77482ff
  kwen2501

  authored
  77482ff

  Copy full SHA for 77482ff

• Change the function signature in trainer to support flexible inputs (#1039)

  Show description for 351e9fb
  wwwjn

  authored
  351e9fb

  Copy full SHA for 351e9fb

• [DeepSeek] remove extra development code (#1051)

  Show description for 7c7923f
  lessw2020

  authored
  7c7923f

  Copy full SHA for 7c7923f

• Supports block causal mask (#1001)

  Show description for 7090a7b
  fegin

  authored
  7090a7b

  Copy full SHA for 7090a7b

• [DeepSeek] Integrate M*G Group Gemm (#1046)

  Show description for 2157f53
  kwen2501

  authored
  2157f53

  Copy full SHA for 2157f53

• improve TrainSpec and general typing (#1048)

  Show description for 73bb964
  lkhphuc

  authored
  73bb964

  Copy full SHA for 73bb964

• [GroupGemm] Fix output dimension of M*G kernel (#1045)

  Show description for 52d2b14
  kwen2501

  authored
  52d2b14

  Copy full SHA for 52d2b14

• refactor model size and flops computation into model args (#1047)

  Show description for d8fc8aa
  tianyu-l

  authored
  d8fc8aa

  Copy full SHA for d8fc8aa

• [WIP] M*G Triton group gemm for MoE training (#967)

  Show description for c8c2326
  lessw2020

  authored
  c8c2326

  Copy full SHA for c8c2326

Commits on Apr 2, 2025

• refactor train_spec.loss_fn to build_loss_fn & chunked loss (#1033)

  Show description for bbb1b61
  

  lkhphuc

  and

  tianyu-l

  authored
  bbb1b61

  Copy full SHA for bbb1b61

Commits on Apr 1, 2025

• Correct self reference in Trainer (#1037)

  Show description for 3ce5616
  iwiwi

  authored
  3ce5616

  Copy full SHA for 3ce5616

Commits on Mar 31, 2025

• Adding OBELICS DataLoader (#663)

  Show description for 3e75bae
  TJ-Solergibert

  authored
  3e75bae

  Copy full SHA for 3e75bae

Commits on Mar 29, 2025

• Fix missing f-string in logging (#1032)

  Show description for ecf26c8
  danielvegamyhre

  authored
  ecf26c8

  Copy full SHA for ecf26c8

Pagination

Previous

Next