Code

Allow layer-wise recompute#18566

Code

Merged

pengwa merged 27 commits intomainmicrosoft/onnxruntime:mainfrom pengwa/add_aggresive_recomputemicrosoft/onnxruntime:pengwa/add_aggresive_recomputeCopy head branch name to clipboardDec 12, 2023

+931-231

Commits

Commits on Nov 23, 2023

• full recompute

  pengwa

  committed
  7164c2f

  Copy full SHA for 7164c2f

• docs

  pengwa

  committed
  7217ad3

  Copy full SHA for 7217ad3

• fix

  pengwa

  committed
  eed2916

  Copy full SHA for eed2916

• minor

  pengwa

  committed
  afc9674

  Copy full SHA for afc9674

• fix memory inspector

  pengwa

  committed
  1df5dd7

  Copy full SHA for 1df5dd7

• move files

  pengwa

  committed
  deedc44

  Copy full SHA for deedc44

• typo

  pengwa

  committed
  15a0640

  Copy full SHA for 15a0640

Commits on Nov 24, 2023

• allow layer-wise recompute

  pengwa

  committed
  4a88196

  Copy full SHA for 4a88196

• fixes

  pengwa

  committed
  f81a3d6

  Copy full SHA for f81a3d6

Commits on Nov 25, 2023

• add layerwise recompute test case

  pengwa

  committed
  d79d171

  Copy full SHA for d79d171

• missing test file

  pengwa

  committed
  c5ba05a

  Copy full SHA for c5ba05a

Commits on Nov 28, 2023

• refine a bit

  pengwa

  committed
  3334d4d

  Copy full SHA for 3334d4d

• fixes

  pengwa

  committed
  bf89c04

  Copy full SHA for bf89c04

• minor

  pengwa

  committed
  38fa66f

  Copy full SHA for 38fa66f

• fix

  pengwa

  committed
  e8817c7

  Copy full SHA for e8817c7

Commits on Nov 30, 2023

• Gelu is followed with a Reshape for gradient ops, but its NodeIndex is even smaller than the other one in forward pass (FusedMatmul which is replaced by a new node after gradient graph is built)

  pengwa

  committed
  cb4eb7b

  Copy full SHA for cb4eb7b

• fix

  pengwa

  committed
  486da22

  Copy full SHA for 486da22

• fix

  pengwa

  committed
  c4e72bb

  Copy full SHA for c4e72bb

• Merge branch 'main' of https://github.com/microsoft/onnxruntime into pengwa/add_aggresive_recompute

  pengwa

  committed
  f3369a8

  Copy full SHA for f3369a8

Commits on Dec 4, 2023

• fix CIs

  pengwa

  committed
  e342f59

  Copy full SHA for e342f59

• Merge branch 'main' of https://github.com/microsoft/onnxruntime into pengwa/add_aggresive_recompute

  pengwa

  committed
  7c3bbc3

  Copy full SHA for 7c3bbc3

• enable tile recompute UT

  pengwa

  committed
  30aebbc

  Copy full SHA for 30aebbc

Commits on Dec 8, 2023

• refine review comments

  pengwa

  committed
  b8e6088

  Copy full SHA for b8e6088

• Merge branch 'main' of https://github.com/microsoft/onnxruntime into pengwa/add_aggresive_recompute

  pengwa

  committed
  0a5f4d4

  Copy full SHA for 0a5f4d4

• lint

  pengwa

  committed
  0fdfa4b

  Copy full SHA for 0fdfa4b

• typos

  pengwa

  committed
  1d0c2bc

  Copy full SHA for 1d0c2bc

Commits on Dec 11, 2023

• typo

  pengwa

  committed
  1602060

  Copy full SHA for 1602060