Code

Bert fix1#87

Code

Draft

goliaro wants to merge 80 commits intoinferenceflexflow/flexflow-serve:inferencefrom bert_fix1flexflow/flexflow-serve:bert_fix1Copy head branch name to clipboard

+3,760-627

Commits

Commits on Apr 17, 2023

• push changes to print dot graph

  

  Teresa Noyola

  committed
  2436152

  Copy full SHA for 2436152

Commits on Apr 21, 2023

• padded input to 512, added isExact to slice_tensor

  

  Teresa Noyola

  committed
  7030ae0

  Copy full SHA for 7030ae0

Commits on May 7, 2023

• Add multiprecision support to replicate

  lockshaw

  committed
  280d29a

  Copy full SHA for 280d29a

• Add explicit template instantiations for replicate kernels

  lockshaw

  committed
  8e84401

  Copy full SHA for 8e84401

Commits on May 8, 2023

• Fix incorrect instantiations

  lockshaw

  committed
  a2ddb91

  Copy full SHA for a2ddb91

• Add nop init_task for replicate

  lockshaw

  committed
  52cc8e8

  Copy full SHA for 52cc8e8

• Fix replicate init_task registration

  lockshaw

  committed
  9eb530a

  Copy full SHA for 9eb530a

Commits on May 10, 2023

• Hopefully print hip errors

  lockshaw

  committed
  d7a219c

  Copy full SHA for d7a219c

• Instantiate extra hip replicate kernels

  lockshaw

  committed
  f323a6e

  Copy full SHA for f323a6e

• fix

  jiazhihao

  committed
  9a75f24

  Copy full SHA for 9a75f24

• Merge branch 'BertMLM_fixes' of https://github.com/flexflow/FlexFlow into BertMLM_fixes

  jiazhihao

  committed
  fe61cbe

  Copy full SHA for fe61cbe

• debug changs

  jiazhihao

  committed
  c9277f3

  Copy full SHA for c9277f3

Commits on May 12, 2023

• Add slice_tensor fix

  

  Teresa Noyola

  committed
  fe66561

  Copy full SHA for fe66561

• Merge branch 'BertMLM_fixes' of github.com:flexflow/FlexFlow into BertMLM_fixes

  

  Teresa Noyola

  committed
  9e04302

  Copy full SHA for 9e04302

• Add logging for metrics

  lockshaw

  committed
  1177748

  Copy full SHA for 1177748

• Add the cuda metrics hack to hip kernel as well

  lockshaw

  committed
  5b7cace

  Copy full SHA for 5b7cace

• Add parallel dim pretty printing

  lockshaw

  committed
  e798a91

  Copy full SHA for e798a91

• [Embedding] bug fix

  jiazhihao

  committed
  90541cf

  Copy full SHA for 90541cf

• Merge branch 'BertMLM_fixes' of https://github.com/flexflow/FlexFlow into BertMLM_fixes

  jiazhihao

  committed
  63fcde6

  Copy full SHA for 63fcde6

• Add replica dim to pretty print

  lockshaw

  committed
  7862143

  Copy full SHA for 7862143

• Merge remote-tracking branch 'refs/remotes/origin/BertMLM_fixes' into BertMLM_fixes

  lockshaw

  committed
  9663d96

  Copy full SHA for 9663d96

• Fix replicate issue with python hack

  tnoyola

  committed
  ef43c36

  Copy full SHA for ef43c36

Commits on May 20, 2023

• Use local json submodule

  lockshaw

  committed
  dd8090e

  Copy full SHA for dd8090e

• ofi conduit-related fixes

  

  Teresa Noyola

  committed
  0dc6187

  Copy full SHA for 0dc6187

Commits on May 22, 2023

• Add mpi flags for hip

  lockshaw

  committed
  0950ac7

  Copy full SHA for 0950ac7

Commits on May 24, 2023

• fix fusion bug

  jiazhihao

  committed
  4b06040

  Copy full SHA for 4b06040

• Merge branch 'BertMLM_fixes' of https://github.com/flexflow/FlexFlow into BertMLM_fixes

  jiazhihao

  committed
  6796b1c

  Copy full SHA for 6796b1c

• increase the max number of regions in a ZeroInitMeta from 64 to 128

  jiazhihao

  committed
  99e9f95

  Copy full SHA for 99e9f95

• support mixed precision

  jiazhihao

  committed
  282c44a

  Copy full SHA for 282c44a

• undo changes to Fused::Transpose

  jiazhihao

  committed
  992dcb9

  Copy full SHA for 992dcb9

• undo changes to config.linux

  jiazhihao

  committed
  f528774

  Copy full SHA for f528774

Commits on Jun 2, 2023

• try to fix layernorm

  jiazhihao

  committed
  a68150d

  Copy full SHA for a68150d

• fix typo

  jiazhihao

  committed
  2bf9afc

  Copy full SHA for 2bf9afc

Commits on Jun 3, 2023

• Add possible layernorm fix

  lockshaw

  committed
  f6f7a32

  Copy full SHA for f6f7a32

• Fix additional layernorm bug due to get_piece_size return size in bytes

  lockshaw

  committed
  5e03b0a

  Copy full SHA for 5e03b0a

• Bugfixes

  tnoyola

  committed
  53fb8bd

  Copy full SHA for 53fb8bd

• Actually check elementwise_affine

  lockshaw

  committed
  449a14c

  Copy full SHA for 449a14c

Commits on Jun 5, 2023

• Revert "Actually check elementwise_affine"

  Show description for c737be6
  tnoyola

  committed
  c737be6

  Copy full SHA for c737be6

Commits on Jun 6, 2023

• Change optimizer to adam with correct hyperparams

  lockshaw

  committed
  a98e09d

  Copy full SHA for a98e09d

• Merge remote-tracking branch 'refs/remotes/origin/BertMLM_fixes' into BertMLM_fixes

  tnoyola

  committed
  66b805e

  Copy full SHA for 66b805e

Commits on Jul 4, 2023

• fix training bert model.

  xinhaoc

  committed
  4bec811

  Copy full SHA for 4bec811

• revert changes

  xinhaoc

  committed
  2d28c15

  Copy full SHA for 2d28c15

Commits on Jul 5, 2023

• fix bert training issue. (#832)

  xinhaoc

  authored
  2025d56

  Copy full SHA for 2025d56

Commits on Jul 25, 2023

• Improve machine_view hash

  lockshaw

  committed
  5f793c1

  Copy full SHA for 5f793c1

• Fix bugs in improved hashing

  lockshaw

  committed
  2c09397

  Copy full SHA for 2c09397

• fix weight dimension in layernorm

  xinhaoc

  committed
  862e9d7

  Copy full SHA for 862e9d7

• Merge branch 'BertMLM_fixes' of https://github.com/flexflow/FlexFlow into BertMLM_fixes

  xinhaoc

  committed
  d29bf1d

  Copy full SHA for d29bf1d

Commits on Aug 8, 2023

• Merge remote-tracking branch 'origin/master' into BertMLM_fixes

  lockshaw

  committed
  88ad5fa

  Copy full SHA for 88ad5fa

Commits on Aug 11, 2023

• fix preregister_task_variant issue, linting

  goliaro

  committed
  2eee875

  Copy full SHA for 2eee875

Commits on Aug 14, 2023

• try to run graph_optimize on each node

  jiazhihao

  committed
  b9d1332

  Copy full SHA for b9d1332

• remove unnecessary file

  jiazhihao

  committed
  b5b0815

  Copy full SHA for b5b0815

Commits on Aug 15, 2023

• fix hip build

  xinhaoc

  committed
  94e35d9

  Copy full SHA for 94e35d9

• Merge branch 'BertMLM_fixes' of https://github.com/flexflow/FlexFlow into BertMLM_fixes

  xinhaoc

  committed
  ac185e3

  Copy full SHA for ac185e3

Commits on Aug 18, 2023

• bypass simulator creation when only_data_parallel is specified

  jiazhihao

  committed
  ded175c

  Copy full SHA for ded175c

• add nccl prints

  jiazhihao

  committed
  1f7e8b7

  Copy full SHA for 1f7e8b7

Commits on Aug 21, 2023

• .

  jiazhihao

  committed
  3fb70f6

  Copy full SHA for 3fb70f6

Commits on Aug 29, 2023

• rccl

  xinhaoc

  committed
  d652b62

  Copy full SHA for d652b62

Commits on Oct 6, 2023

• fix fuse

  xinhaoc

  committed
  b39528b

  Copy full SHA for b39528b

• fix hip

  xinhaoc

  committed
  0cf3c8e

  Copy full SHA for 0cf3c8e

• more fix to hip

  xinhaoc

  committed
  17a1c4e

  Copy full SHA for 17a1c4e

Commits on Nov 3, 2023

• customized kernel for broadcasting add.

  xinhaoc

  committed
  f65044d

  Copy full SHA for f65044d

Commits on Dec 20, 2023

• dropout

  xinhaoc

  committed
  bcab56a

  Copy full SHA for bcab56a

• optimizer

  xinhaoc

  committed
  fa1fffc

  Copy full SHA for fa1fffc

Commits on Dec 21, 2023

• opt

  xinhaoc

  committed
  40d830c

  Copy full SHA for 40d830c

• fix

  xinhaoc

  committed
  e825526

  Copy full SHA for e825526

Commits on Jan 12, 2024

• fix

  xinhaoc

  committed
  d2bdb15

  Copy full SHA for d2bdb15

• .

  xinhaoc

  committed
  3b9e1c6

  Copy full SHA for 3b9e1c6

• fix

  xinhaoc

  committed
  9f8bb9e

  Copy full SHA for 9f8bb9e

• remove print

  xinhaoc

  committed
  fb91122

  Copy full SHA for fb91122

Commits on Jan 19, 2024

• fix hip

  xinhaoc

  committed
  c162d4c

  Copy full SHA for c162d4c

• fix multinodes

  xinhaoc

  committed
  ea79317

  Copy full SHA for ea79317

• fix

  xinhaoc

  committed
  58d84ed

  Copy full SHA for 58d84ed

Commits on Jan 25, 2024

• fix

  xinhaoc

  committed
  01c9d4c

  Copy full SHA for 01c9d4c

Commits on Jan 26, 2024

• fix

  xinhaoc

  committed
  a31f8e9

  Copy full SHA for a31f8e9

Commits on Feb 2, 2024

• tp

  xinhaoc

  committed
  9141c46

  Copy full SHA for 9141c46

Commits on Feb 15, 2024

• timer

  xinhaoc

  committed
  8185289

  Copy full SHA for 8185289

Commits on Feb 16, 2024

• rmv

  xinhaoc

  committed
  d958805

  Copy full SHA for d958805

Commits on Feb 22, 2024

• fix tp

  xinhaoc

  committed
  38dfd87

  Copy full SHA for 38dfd87

Commits on Feb 29, 2024

• try a fix

  xinhaoc

  committed
  355d4b4

  Copy full SHA for 355d4b4

Commits on Sep 13, 2024

• fix hip

  xinhaoc

  committed
  8488ba0

  Copy full SHA for 8488ba0