Code

Gradient Fuse Allreduce#45643

Code

Merged

JZ-LIANG merged 20 commits intoPaddlePaddle:developPaddlePaddle/Paddle:developfrom JZ-LIANG:AutoParallel/dp_allreduce_fuseJZ-LIANG/Paddle:AutoParallel/dp_allreduce_fuseCopy head branch name to clipboardSep 14, 2022

+319-39

Commits

Commits on Aug 24, 2022

• bugfix (PaddlePaddle#45332)

  JZ-LIANG

  committed
  b1007f2

  Copy full SHA for b1007f2

• dist embedding support lookup table v1

  JZ-LIANG

  committed
  a3e4a2f

  Copy full SHA for a3e4a2f

Commits on Aug 25, 2022

• add unitest

  JZ-LIANG

  committed
  27b50e8

  Copy full SHA for 27b50e8

• Merge remote-tracking branch 'upstream/develop' into AutoParallel/dist-embedding-support-3d-input

  JZ-LIANG

  committed
  7a1662a

  Copy full SHA for 7a1662a

Commits on Aug 26, 2022

• customize wait_comm

  JZ-LIANG

  committed
  ba35eeb

  Copy full SHA for ba35eeb

Commits on Sep 1, 2022

• group gradients

  JZ-LIANG

  committed
  ea24b07

  Copy full SHA for ea24b07

• Merge remote-tracking branch 'upstream/develop' into AutoParallel/dp_overlap_support_weight_sharding

  JZ-LIANG

  committed
  bc9e802

  Copy full SHA for bc9e802

• Merge branch 'AutoParallel/dp_overlap_support_weight_sharding' into AutoParallel/dp_allreduce_fuse

  JZ-LIANG

  committed
  5577126

  Copy full SHA for 5577126

• bugfix

  JZ-LIANG

  committed
  13e31b5

  Copy full SHA for 13e31b5

Commits on Sep 2, 2022

• update program

  JZ-LIANG

  committed
  8ca8a38

  Copy full SHA for 8ca8a38

Commits on Sep 6, 2022

• disable when sharding

  JZ-LIANG

  committed
  06f467d

  Copy full SHA for 06f467d

• engine with profile

  JZ-LIANG

  committed
  2fb0e28

  Copy full SHA for 2fb0e28

Commits on Sep 7, 2022

• improve recompute ckpts

  JZ-LIANG

  committed
  04ec849

  Copy full SHA for 04ec849

Commits on Sep 9, 2022

• update fp16 pass

  JZ-LIANG

  committed
  3f5207b

  Copy full SHA for 3f5207b

• bugfix

  JZ-LIANG

  committed
  8d148c2

  Copy full SHA for 8d148c2

• bugfix

  JZ-LIANG

  committed
  af1c77d

  Copy full SHA for af1c77d

• bugfix

  JZ-LIANG

  committed
  02d2057

  Copy full SHA for 02d2057

Commits on Sep 13, 2022

• remvoe local changed

  JZ-LIANG

  committed
  3c07d13

  Copy full SHA for 3c07d13

• Merge remote-tracking branch 'upstream/develop' into AutoParallel/dp_allreduce_fuse

  JZ-LIANG

  committed
  ff6c597

  Copy full SHA for ff6c597

• enable fuse

  JZ-LIANG

  committed
  5aee3f1

  Copy full SHA for 5aee3f1